A health promotion intervention to improve lifestyle choices and health outcomes in people with psychosis: a research programme including the IMPaCT RCT

Cannabis use

Over the last two decades, there has been a justified interest in comorbid substance use by those with a psychotic illness because of its high prevalence and significant impact on clinical and social problems, as well as the heavy burden laid on the health services. According to the Health and Social Care Information Centre’s 2014 report36 on statistics in drug misuse in England, primary diagnoses of a drug-related mental health and behavioural disorder increased by 8.5% from 2012/13 to 2013/14. 36 The size of the problem becomes even more important when those with a psychotic illness use substances significantly more than the general population.

The interaction between a psychotic illness and substance use is complex and can have major detrimental effects on the course of the illness, including a patient’s risk of experiencing violence and being hospitalised, their ability to comply with treatment and even possibly the aetiology. 37,38 Such patients may also be at particular risk of deterioration in their mental health. A recent systematic meta-analysis of the outcomes associated with psychosis and comorbid substance use showed that current substance users with psychosis may have more severe positive symptoms than patients who have never used substances. 39

Aetiological links between cannabis and psychosis

Interestingly, cannabis was used 150 years ago to treat ‘insanity’, while also being recognised as increasing the risk of ‘madness’, especially among young persons at the Maudsley Hospital. 40 Apart from some experimental studies examining the effects of cannabis during the 1960s, no interest was shown in cannabis studies. In 1987, Andréasson et al. 41 published a longitudinal study involving > 50,000 Swedish conscripts and reported that those who used cannabis by the age of 18 years were twice as likely to develop schizophrenia as non-users. The risk increased to sixfold in heavy cannabis users. This study was criticised and ignored until the early 2000s, when Zammit et al. 42 reanalysed the Swedish data, coming up with similar findings. The subject of cannabis and psychosis once more attracted empirical interest and the subsequent publications indicated a causal link. 43–45 Other studies, including systematic reviews, concluded that using cannabis increased the risk of developing a psychotic illness by two to six times, particularly in those with a predisposition to the condition and in a dose-dependent manner. 37,46,47

Furthermore, experimental studies carried out on healthy volunteers have shown that cannabis can induce transient psychosis in some individuals, but not all. 48–50 At the same time, biochemical, imaging and genetic studies were carried out to examine the links between cannabis and psychosis, which have all added to the existing knowledge. For instance, we now know that the main psychoactive ingredient of cannabis, delta-9-tetrahydrocannabinol (THC), binds to CB1 receptors within the endocannabinoid system, and the very same brain regions are also implicated in psychoses, particularly in schizophrenia. 51–53

Cannabis use in first episode of psychosis patients, prevalence and increased risks

Cannabis is the most frequently used illicit substance in the world, with nearly half of 15- to 34-year-olds reported to be lifetime users in the USA, Australia and Canada. 54 According to the 2015 European Monitoring Centre for Drugs and Drug Addiction (EMCDDA) report,55 lifetime use of cannabis among the same age group in the UK is reported to be 29.9%, whereas the last 12-month use was 11.2%. Cannabis is also the most commonly used substance after tobacco, according to epidemiological surveys of people with psychosis. 56–59 Perhaps unsurprisingly, we found that the use of substances other than cannabis among first episode of psychosis (FEP) patients was minimal. Given the changing attitudes towards cannabis use, reflected by its ever-increasing medicinal use or decriminalisation in some countries, its more widespread use by new users is expected. 60 It is, therefore, important that the health risks of cannabis use are sufficiently researched.

At the time of the original National Institute for Health Research (NIHR) Programme Grants for Applied Research application, we wrote that a higher percentage of people with psychosis used cannabis and other drugs than the general population61 and regular cannabis use increased the risk of schizophrenia two- to threefold. 44 We also highlighted that ongoing cannabis use in those with psychosis led to increased relapse and hospitalisation, lack of compliance with treatment and longer illness. 39

Indeed, a considerable number of studies since have further emphasised the impact of cannabis use on people with psychosis. First, the higher prevalence of cannabis use among people with FEP has been validated in numerous studies. 62,63 A recent meta-analysis64 on the prevalence of cannabis use in people with FEP pooled the data from 37 studies and concluded that cannabis use is highly prevalent in this group. The same study64 also reported that the pooled estimate from 10 studies showed that regular use of cannabis begins 6.3 years before the onset of psychosis. In fact, the same research group has previously shown in another meta-analysis65 that the age at the onset of psychosis among cannabis users was nearly 3 years earlier.

Additionally, prospective epidemiological studies and meta-analyses of such studies have concluded that using cannabis can indeed increase the risk of developing a psychotic illness and that there is a dose-related risk. 37,66–68 It should be noted that the main psychoactive compound of cannabis (i.e. THC) is nearly three times as concentrated in the cannabis used today as that used in earlier decades.

In a study in South East London, using ‘skunk-like’ cannabis was shown to triple the risk of psychosis and, when the results were extrapolated across the population, the use of high-potency cannabis was found to be associated with a 24% increase in new cases of psychosis. 67 It was also shown that both the frequency of use and the strength of the variety of cannabis used were associated with the increased risk. In another recent study with FEP patients carried out across Europe and Brazil in 11 sites, daily cannabis use with high-potency varieties increased the odds of psychotic disorder nearly five times in comparison with never users, and population attributable fractions calculations indicated that, if such high-THC cannabis had not been available, 12.2% of cases of FEP could have been prevented across these sites. 69

Why do first episode of psychosis patients use cannabis?

The questions of why people with FEP use cannabis significantly more than others and why some continue to use it despite its deleterious effects remain important and have been explored using various theories. One of the first proposed theories was ‘self-medication’ hypothesis or ‘reverse causality’, suggesting that psychiatric symptoms are alleviated by cannabis or that cannabis is used to counter the side effects of psychotropic medication. However, this theory has since been disputed by several studies. 46,47,70 The recent meta-analysis64 finding of a 6.3-year gap between the initiation of cannabis use and the onset of FEP is further evidence against this proposal.

However, some studies support the self-medication or reverse causality theory, as it is possible that some early dysphoric symptoms or anxiety make it more likely that some people will use cannabis. One such study carried out a comprehensive examination of experiences of cannabis use and reasons for using cannabis among people with psychosis and found that the primary reason given was not related to positive or negative symptoms or side effects of medication,71 as only 10% reported using it for these reasons. Instead, the most frequently cited motivations were to reduce boredom, to improve socialisation and to alleviate some symptoms, such as agitation (47%) and difficulty sleeping (43%). Participants also reported that cannabis reduced feelings of depression.

The increased use of cannabis among people with psychosis has also been attributed to certain personality traits/disorders. There is mounting evidence that high-scoring schizotypal people who use cannabis are more likely to experience psychotic symptoms. 72–74 Another study has shown that conduct disorder symptoms are significantly associated with use of cannabis, particularly if it is used by the age of 14 years, among people with FEP; conduct disorder symptoms have been suggested to independently increase the possibility of cannabis use, which then increases the risk of psychosis. 75

Finally, genetic studies have gained particular significance because most cannabis users do not develop psychosis; therefore, genetic factors need to be examined to find out what determines sensitivity to the psychosis-inducing effects of cannabis. 76 One of the earlier theories was genetic confounding: in other words, that cannabis use and psychosis risk shared genetic origins. However, more recently gene–environment interaction studies have gained more recognition; these refer to an environmental factor, such as the use of cannabis, being influenced by genetic factors. For instance, when there was a familial risk, taken as the measure of genetic loading, the non-psychotic siblings of people with psychosis were found to be significantly more vulnerable to mental health disturbances than the control participants. 77 Further support was provided by another study that showed increased sensitivity to the psychotogenic effects of cannabis to be associated with familial risk of psychosis. 78

In addition to familial risk factors, there have been a number of candidate gene studies. The first of these was carried out by Caspi et al.,79 who focused on a functional polymorphism of the valine allele of catechol-O-amyltransferase (COMT) gene in the New Zealand birth cohort study in examining the interaction between cannabis use and risk of psychosis. Later, a number of studies concentrated on this gene, leading to mixed results when further replication and validation studies were carried out. 80,81 However, one promising gene candidate is thought to be alpha serine/threonine-protein kinase (AKT1) as a variant of this gene displayed a twofold increased risk of a psychotic disorder after use of cannabis. 82 This finding was later independently replicated. 83 However, further studies need to be carried out because of the small evidence base.

In conclusion, the available evidence suggests that the interaction between the increased use of cannabis and the development of a psychotic illness is complex and may have both environmental and genetic causes.

What are the physical health effects of cannabis on first episode of psychosis patients?

Although the mental health risks of cannabis have been well studied, there is paucity of research on its possible health-care risks not only in psychiatric but also in general populations.

The question of why cannabis could affect physical health is linked to the knowledge that the endocannabinoid system where THC binds exists not only in the brain but also in most other organs, such as the heart, lungs, liver, kidneys, thyroid, bones and reproductive organs, as well as the immune system. 84–86

Although knowledge about the multiple functions of this widespread neurotransmitter system is evolving, the available evidence implies that the endocannabinoid system has varied roles ranging from regulating the metabolism, circulatory system, reproduction, sleep and pain to ocular pressure. 87–90 It would therefore be possible to assume that the interference of THC, by interrupting the normal functioning of this widespread system, may lead to physical health problems. Only during the last few years has there been an interest in studying the physical health-care effects of cannabis among non-psychiatric populations.

One possible risk is to the functioning of the respiratory system as a result of the fact that cannabis is usually inhaled. The use of both cannabis and tobacco is common,91 which also creates confounding effect problems in research. Findings so far show that there may be a dose-related risk, in that low levels of cannabis use (three to five joints per month) may increase respiratory function, whereas higher levels of use have the opposite effect. 59,92 So far, a conclusive association has not been established between cannabis use and lung cancer. 93

The endocannabinoid system plays an important role in the regulation of food intake and reduces energy expenditure. 94,95 Based on this knowledge, some cannabis compounds are used to stimulate appetite and encourage weight gain in some patients with HIV (human immunodeficiency virus), AIDS (acquired immunodeficiency syndrome) or cancer. Interestingly, however, some recent studies have shown that in the general population cannabis use is associated with lower body mass than in non-users. 96 This paradoxical finding may be due to various factors. For instance, there may be a difference between long-term use, as found in general population studies, and short-term use, as found in appetite stimulation studies. The other explanation would be the possibility of poly-drug use or food and drugs competing for the same reward sites in the brain. It is also suggested that cannabis may act as a regulatory compound, increasing weight in those with low weight but not in those who are already overweight. 97 Interestingly, however, most recent studies carried out not only on the general population98–100 but also on people with SMI101 report that cannabis use is associated with lower BMI, smaller waist circumference, lower diastolic BP and more severe psychotic symptoms.

Unfortunately, no established and practical treatment programmes target both improving physical health and reducing substance use among people with psychosis, and attending parallel treatment programmes is often impractical. Treatment programmes, such as MIDAS, for people with both psychosis and substance misuse appeared promising,102,103 but these are lengthy, complex and expensive.

The huge implications of these health gaps in terms of morbidity, mortality, QoL and projected future cost make addressing physical health in SMI a national priority. The National Service Framework and the National Institute for Health and Care Excellence (NICE) have emphasised the importance of good physical health in people with SMI and encourage primary and secondary care services to collaborate to improve physical outcomes in this population. 104,105 In 2004, the UK government emphased physical health promotion as a means of reducing CVD burden and endorsed the aims of programmes addressing physical health in people with SMI. 106 None of these programmes, however, was adequately tested in the UK to ensure that it could accurately quantify the problem in those with SMI, could identify those most at risk, could reduce physical health risk factors and was reliable and reproducible enough to be disseminated across the NHS.

This is a matter of equity. A decade ago, the UK Disability Rights Commission investigation stated overtly that people with mental health problems were more likely than others to experience major illness and develop serious health conditions earlier to and die earlier as a result. 107 Yet this group were less likely to receive some important treatments/health checks and faced real barriers to accessing services. The report called for a clear shift in approach to eliminate inequitable treatment and to target high-risk groups. The idea was that this would prevent the extra costs of serious ill health being passed on to other parts of the NHS and enable people with SMI to be healthier and participate fully in society. 107 Similar concerns were reflected in the Choosing Health: Making Healthy Choices Easier White Paper. 106 Since then, this topic has gathered momentum; there have been a number of high-profile policy documents, including No Health Without Mental Health,108 the Department of Health and Social Care report Closing The Gap: Priorities For Essential Change in Mental Health,109 Whole-person Care: from Rhetoric to Reality,110 the BMA report Recognising the Importance of Physical Health in Mental Health and Intellectual Disability,111 the Annual Report of the Chief Medical Officer 2013112 and the London Health Commission report Better Health for London. 113 Indeed, the London Health Commission’s ambition to reduce the gap in life expectancy between adults living with SMI and the rest of the population has led to the creation of a working group, ‘Stolen Years’, to work across health systems to make this a reality.

The challenge remains the evidence base. Most studies of health promotion in SMI to date have focused on selected groups of patients and achieved only modest outcomes. None has addressed comorbid substance misuse. Importantly, none has evaluated the cost-effectiveness of a combined lifestyle intervention across a service, work needed to inform health commissioning. The next logical step was to develop a more practical alternative to separate programmes, and so we set out to develop and evaluate the benefits of a well-defined, standardised, person-centred intervention, using modules, as appropriate, to target both lifestyle and substance use to maximise physical and mental health. We also undertook an evaluation of the costs of the cardiovascular burden and the cost-effectiveness of the intervention to reduce it.

Healthy living interventions should be an integral part of the care provided to people with SMI. This must become a priority for both primary and secondary care services, although it will require a cultural shift in terms of how care is provided to those with SMI, with a greater emphasis on a holistic approach. There is an urgent need for (1) a reliable way of identifying those most at risk, (2) guidelines on how best to screen for the emergence of MetS and (3) an effective intervention to reduce modifiable cardiovascular risk factors in this group.

Work package 1: PUMP

FIGURE 1.

Research design pathway.

The Physical health and substance Use Measures in first episode of Psychosis (PUMP) study was a prospective observational cohort of patients presenting with their FEP. The participants were recruited from the four boroughs of the SLaM NHS Foundation Trust, two boroughs of the Oxleas NHS Foundation Trust and two boroughs of the Sussex Partnership NHS Foundation Trust. Patients from both inpatient and outpatient services were followed up for 12 months to observe health and service utilisation markers over time and to test the association between lifestyle (diet, exercise and substance use) and antipsychotic medication factors and the emergence of components of MetS.

In the first part of the IMPaCT programme, we also wanted to examine the extent of cannabis use among our group of FEP patients, as well as finding out the mental and physical health-care effects of this use at baseline and at 12 months’ follow-up, compared with non-users.

Work package 2: development of the health promotion intervention (IMPaCT therapy)

The available evidence was used to inform the development, validation and manualisation of a health promotion intervention (HPI) tailored to the needs of the individual, addressing a number of lifestyle choices, including substance use. The purpose of the development of the manual was to create a culturally appropriate, innovative and effective programme for use in routine care systems to achieve better physical and mental health in people with SMI by improving lifestyle choices and decreasing substance use. This intensive HPI (IMPaCT therapy) was designed to cover physical health, mental health and substance use using motivational interviewing (MI) and cognitive–behavioural therapy (CBT) and was designed to be sufficiently pragmatic to be deliverable within the NHS.

Work package 3: IMPaCT randomised controlled trial

A randomised controlled trial (RCT) assessing whether or not the addition of an intensive HPI (IMPaCT therapy) to usual mental health care delivered by care co-ordinators is more effective and more cost-effective than usual mental health care in improving metabolic outcomes and reducing substance use among people with SMI on completion of the intervention at 12 months and 15 months after baseline.

Measures

A range of measures were collected over three time points (baseline and 3 and 12 months), including the Positive and Negative Syndrome Scale (PANSS), Clinical Global Impression Severity scale, Global Assessment of Functioning (GAF), Young Mania Rating Scale, Calgary Depression Scale and operational criteria checklist for psychotic and affective illness (OPCRIT) diagnostic scale, as well as anthropometric measures and blood sampling for cardiometabolic and inflammatory markers and measures of nicotine, alcohol and substance use, including the Nicotine Dependence questionnaire (Fagerström scale), the Alcohol Use Disorders Identification Test (AUDIT) and a modified Cannabis Experience Questionnaire (CEQ-4), along with the International Physical Activity Questionnaire (IPAQ). 72,115–121 All measures in the study were administered face to face by a trained researcher, except the OPCRIT, which was rated by a clinician based on the clinical notes.

Follow-up contacts

Clinical, sociodemographic and substance use data were also collected using the Follow-up Psychiatric and Personal History Schedule (FUPPHS). This records information about the patient’s mental state, general behaviour, substance use, events and personal history over a defined period using information obtained from patients, informants, case notes and other records. 122 Researchers rating the Psychiatric and Personal History Schedule (PPHS) at follow-up achieved an intraclass correlation of 0.90 on all PPHS items when duplicate ratings were compared. Information on medication adherence and remission during the 1-year follow-up was also extracted from the PPHS, where poor adherence is defined as 1 = lapses of ≥ 3 days more than once, and 2 = not taking any prescribed medication. Remission was operationally defined as the absence of positive, negative or disorganised symptoms for at least 30 days. In our study122 examining the effect of substance use and medication adherence on outcomes, the 1-year follow-up period was taken as the date of first contact with mental health services of the SLaM NHS Foundation Trust for psychosis to the date exactly 1 year later using the clinical records held on the SLaM electronic Patient Journey System (ePJS), with all of the PPHS measures completed by a researcher retrospectively using the electronic mental health records system (i.e. the SLaM electronic Patient Journey System).

Linear regression models were used to examine for the associations between baseline diet, alcohol use, sedentary behaviour and prescriptions of olanzapine with cardiometabolic risk both at baseline and at 12 months’ follow-up. Baseline cross-sectional associations between continuous scores for lifestyle and cardiometabolic factors were investigated using unadjusted and adjusted linear regression models, as was the relationship between baseline lifestyle choices and change in cardiometabolic factors at 12 months, investigating associations between use of dibenzodiazepine medications between baseline and 12 months (yes/no) and cardiometabolic factors in the same way. We used the mi impute chained command in Stata^® version 15.1 (StataCorp LP, College Station, TX, USA) with 50 imputed data sets for each MI model to account for different patterns of missingness. To account for multiple testing, a stricter alpha of 0.01 was prespecified as the significance level. The assumptions of the regression analyses of constant variance and normal distribution were assessed with a visual inspection of residual plots.

We used univariate and multivariate analyses to determine the relationship between measures of cardiometabolic risk at baseline and admission, and home treatment costs in the following year (see Appendix 1).

Patient and public involvement

Service user and carer experts were part of the core planning and management throughout the project (see Programme patient and public involvement).

Metabolic characteristics and changes over 12 months

There were significant levels of cardiometabolic risk at first presentation, with rates of obesity, raised levels of glycated haemoglobin (HbA_1c), levels of C-reactive protein (CRP) and low levels of high-density lipoprotein (HDL) cholesterol increasing over the following 12 months. In the overall group, central obesity rates at baseline were higher in women (62.7%; n = 37/59; waist circumference ≥ 80 cm) than in men (35.3%; n = 36/102; waist circumference ≥ 94 cm; χ² = 11.34; p = 0.001). The mean waist circumference in white men increased by 4.9 cm over the year, such that by 12 months white men had a waist measurement 7.3 cm greater than their counterparts of other ethnicities, despite having comparable measures at baseline. The average waist circumference of white women at baseline was 9.6 cm smaller than of women of other ethnicities.

The mean total cholesterol levels among those participants of white ethnicity (4.9 mmol/l) and those of other ethnicities (4.7 mmol/l) were comparable at baseline, but white patients had higher levels at 12 months (5.1 vs. 4.6 mmol/l; p = 0.04) (see Gaughran et al. 123).

Baseline antipsychotic use and associated metabolic impact

The median duration of treatment with antipsychotics at baseline was 21 days (interquartile range 9–55.5 days) (mean 33.7 days, SD 50.3 days), with 95% of the sample prescribed second-generation antipsychotics. The most frequently prescribed antipsychotic was olanzapine.

Patients who had been prescribed antipsychotic medication for ≥ 2 weeks had an average total cholesterol higher (0.5 mmol/l, 95% CI 0.1 to 0.8 mmol/l; p = 0.007) than those who had been prescribed antipsychotics for less than a fortnight. Participants medicated prior to baseline had a higher baseline average waist circumference (7.7 cm, 95% CI 2.0 to 13.4 cm; p = 0.009) than antipsychotic-naive patients.

Lifestyle choices and baseline cardiometabolic outcomes and changes over time

There was no association between any of the baseline or 12-month cardiometabolic outcomes and Dietary Instrument for Nutrition Education (DINE) fat scores, sedentary behaviour, AUDIT scores and pre-baseline olanzapine. The lack of association remained when adjusting for potential confounders (age, gender, ethnicity or pre-baseline days on medication) and in sensitivity analyses using all non-missing data from the full data set.

Subgroup analyses were run comparing those participants who took olanzapine between baseline and 12 months and those who did not. No differences were found between the two groups in the magnitude of associations between changes in cardiometabolic outcomes at 12 months’ follow-up and baseline lifestyle factors, nor were there any such associations within either of the two groups. No associations emerged from sensitivity analyses using the full data set (both unadjusted and adjusted for the same potential confounders) (see Gaughran et al. 123).

Obesity at baseline was associated with higher subsequent admission costs, low HDL cholesterol at baseline was associated with lower subsequent admission costs and higher HDL cholesterol at baseline was associated with a greater subsequent quality-adjusted life-year (QALY) gain (see Appendix 1).

Cannabis use in PUMP first episode of psychosis patients

Of 206 patients with FEP, 102 used cannabis at baseline (49.5%), with 23 out of 183 (12.6%) reporting current use of other recreational drugs. Current users at baseline were more likely to be men [59.4%, degrees of freedom (df) 1; p < 0.05] and most were younger, belonging to the 18–33 years age group (df 1; p < 0.05), than the non-users. They were also more likely to be single (df 4; p < 0.05). In terms of ethnicity, occupation, having close confidants, ICD-10 diagnoses from OPCRIT, drug naivety and seeing GP or health specialists, there were no significant differences between cannabis users and non-users. Cannabis users also differed from non-users in terms of using cigarettes (df 1; p < 0.0001) and other stimulants (df 3; p < 0.01). At 12 months’ follow-up, we had data on 105 patients and the number of current users dropped to 40 (38.1%). The corresponding figure at 12 months for those reporting use of other substances was 12 out of 102 (11.8%). From first data collection to 12 months, 12 (11.3%) participants started using cannabis, whereas 16 people (15.5%) stopped.

Effects of cannabis on outcomes in first episode of psychosis patients

In terms of physical health outcomes, non-cannabis users had higher cholesterol levels at 12 months’ follow-up than non-users (df 1; p <.045). Female users were thinner at baseline (df 1; p < 0.036) but male users were heavier at 12 months (df 1; p < 0.003) than non-users. Female users had smaller waistlines at baseline (df 1; p < 0.008) than female non-users and male users had larger waistlines at 12 months (df 1; p < 0.01) than male non-users.

There has been much speculation about why people with a diagnosis of psychosis may use cannabis; some suggest a self-medication hypothesis, whereas other evidence suggests that it is used to alleviate dysphoria. Kolliakou et al. 124 conducted a study nested within the PUMP study, which investigated the reasons for use reported by those participants who used cannabis and how these varied over time. Participants rated their motives at baseline (n = 69), at 3 months (n = 29) and at 12 months (n = 36) on the Reasons for Use Scale, which has five subscales (enhancement, social motive, coping with unpleasant affect, conformity and acceptance, and relief of positive symptoms and side effects). At all of the time points, ‘enhancement’ received most endorsement, followed by ‘coping with unpleasant affect’ and ‘social motive’. ‘Conformity and acceptance’ followed closely. The least endorsed motive was ‘relief of positive symptoms and side effects’. When participants continued to use cannabis at 3 months and 12 months, they endorsed these reasons less strongly than at baseline. We found little support for the theory that people were using cannabis for self-medication or to alleviate dysphoria. Rather, in keeping with the general population, the most common reason that people with their FEP gave for their use of cannabis was ‘enhancement’.

Colizzi et al. 122 demonstrated that substance use and poor medication adherence contribute to poor outcomes in the year following the FEP. Colizzi et al.’s work detailed the relative contributions of medication adherence and substance use to outcome over the year after the FEP. We had data on 205 patients on use of tobacco, alcohol, cannabis and stimulants at psychosis onset and at 1 year, along with data on medication adherence and symptom remission. Rates of overall substance use were high both before (37–65%) and after psychosis onset (45–66%). Nicotine dependence was reported in 53.2% of patients at baseline, whereas 40.5% reported premorbid problem drinking, 65.4% reported premorbid cannabis use and 36.6% reported premorbid stimulant use. A total of 44% of patients had poor medication adherence and 55% failed to reach remission from psychosis.

Both nicotine dependence and cannabis use after the onset of psychosis predicted poor medication adherence and non-remission significantly. Poor medication adherence occurred in 43.9% of patients at some point in the first year and also significantly predicted remission during the 1-year follow-up, with patients with poor medication adherence having a sixfold increased likelihood of psychosis non-remission when compared with those with good adherence. This association between medication adherence and remission was still significant when substance use in the 1-year follow-up period was added into this model. Sobel tests for mediation showed that medication adherence was a significant mediator of the relationship between nicotine dependence and remission (z = 2.02; p = 0.04) and of that between cannabis use and remission (z = 2.12; p = 0.03). 122

Design

Initially, a consultation among experts in therapeutic interventions for mental and physical health, substance use and diabetes generated the key features and focus of the intervention. After this, IMPaCT therapy was developed through three stages of design and analysis: (1) therapy and training development, and manual writing; (2) piloting, evaluation and refining the training package with clinicians; and (3) a Delphi process to reach consensus on the therapy model and manual through initial consultation followed by two rounds of follow-up questionnaire feedback. The feedback was from each of three expert groups: (1) therapists, (2) clinician providers within CMHTs and (3) psychosis service users. The emphasis was on an iterative process leading to a therapy model and manual that would be well grounded both in theory and in feedback from experts in behavioural change therapies, clinician providers within CMHTs and psychosis service users.

The intervention was based on the transtheoretical behaviour change model. 170 The training and reference guide introduced this model of behaviour change and its link with the intervention approach. The intervention manual was then framed around this model. The manual was divided into separate sections for the distinct stages of behaviour change: (1) precontemplation, (2) contemplation, (3) preparation and contingency planning, (4) action and (5) maintenance and relapse prevention. At the start of each section was an overview that defined the stage and how to identify whether or not a client was in this stage. This was followed in each section by aims, objectives and between 3 and 12 specifically behaviour change ‘interventions’ that were consistent with the behaviour change taxonomy171,172 and applicable to that stage.

Participants

Participants in the phase 2 training development were eight staff selected from a range of in- and outpatient settings and backgrounds, working in the area of psychosis within the SLaM NHS Foundation Trust, but not eligible to participate in the subsequent IMPaCT RCT. Participants in the phase 3 Delphi consensus and manual development comprised three expert groups of (1) therapist/researchers, (2) clinicians and (3) service users, each of whom took part in two iterative review and feedback sessions. Therapist/researcher experts were recruited from the local and national community and were involved in the development, implementation or research evaluation of CBT and MI approaches in psychosis. Clinicians were staff who took part in the phase 3 training, and service users were people who were receiving a service from the relevant clinicians at the time and had provided informed consent to participate in the study.

Pilot training

All participants completed a 5-day pilot training, which comprised 10 sessions: (1) introduction to IMPaCT and the manual/basic CBT skills, (2) introduction to MI, (3) practical MI, (4) intermediate MI skills in psychosis I, (5) intermediate MI skills in psychosis II, (6) physical health awareness and monitoring, (7) physical health awareness and diabetes monitoring, (8) substance use awareness, (9) group work skills and (10) healthy living groups. The training was delivered by members of the research team (KG, SS, ZA and Manyara Mushore), as well as an expert in healthy living groups and an expert in MI for psychosis. All participants in the training self-rated of their knowledge and skills before and after the training, and their confidence in therapy delivery after training.

Delphi consultation

The Delphi process comprised three iterative stages of consultation revolving around an adapted questionnaire that incorporated feedback from each preceding stage, consistent with recommended Delphi methodology. The Delphi consultation on the model and manuals was collected in semistructured e-mail questionnaires for therapist/researchers and clinicians, and in face-to-face interviews with service users, in accordance with the algorithm presented by Jones and Hunter173 (see Appendix 2, Figure 2). The Delphi process allowed us to access the acceptability and face validity of the model, manuals and training package, using expert consensus. The Delphi process was an in-depth consultation requiring the expert clinician and patient consultants to undertake a selected practice intervention session in pairs, using the intervention, reference guide, manual and handbook, and provide iterative feedback. Feedback was obtained in the form of qualitative comments and Likert scale ratings. For each feedback session, each participant read specified sections of the manuals. Therapists/researchers read all sections of all manuals (the reference guide, the manual and the Better Health Handbook), whereas clinicians focused preferentially on the therapy manual and handbook sections. The service users took part in two selected intervention sessions with their trained clinician and reviewed preferentially the service user resource handbook. Questions considered user-friendliness, spirit of MI, integration with CBT, usefulness in routine NHS practice, length and complexity, and whether anything was missing or should be added. Feedback was obtained from numerical ratings on a Likert scale (0–10), where 0 was least positive and 10 was most positive, as well as qualitatively. After each round of feedback, changes were made to the content and structure of the manual before the next round of feedback. In follow-up consultations, the previous individual ratings and group rating were also fed back, and participants were encouraged to revise their ratings towards or away from the group mean to determine consensus within each group. Consensus across the group as a whole was based on threshold ratings for the acceptability and user-friendliness of key elements of the manual on Likert scales of ≥ 7 in the final stage.

Alongside the Delphi process, an informal presentation and consultation were conducted with a carers group for people with SMI. The perspectives gained from this presentation were used in the development of a brief carers’ section to support the therapy.

Initial consultation to form key features and components of the intervention and to consider mechanisms of change

A consultation among experts in the field of MI and psychosocial interventions for physical health, substance use and diabetes and CBT for psychosis (KG, SS, ZA, KI and Gill Todd) yielded a set of broad features and foci of the intervention for consideration in subsequent stages of development. The initial therapy built on an adapted version of the physical health intervention used in the Well-Being Support Programme,140,174 incorporating a substance use intervention based on ‘Managing Mental Health and Drug Use’. 142 Key features of the therapy were that it had to be comprehensive, integrative, flexible and pragmatic, deliverable by mental health clinicians from a range of backgrounds, and able to be readily implemented in NHS settings. The focus of the components incorporated (1) mental health and psychosis, (2) alcohol use, (3) cannabis use, (4) smoking, (5) other drug use, (6) diabetes, (7) exercise and (8) healthy eating. It was initially proposed that, based on previous research, the intervention should be delivered in a group format.

Phase 1: initial therapy training development and manual writing

Extensive review of the Well-Being Support Programme and substance use intervention, as well as of relevant literature about mental health, physical health and substance use interventions incorporating CBT and/or MI in psychosis and non-psychosis populations, led to the development of a preliminary outline of the intervention and of the training package.

Phase 2: piloting, evaluation and refining the training package

The pilot training was delivered to clinicians (n = 8; 50% male) from CMHTs (n = 2), to specialist inpatient (n = 3) and psychiatric intensive care units (n = 3), and to individuals from a range of disciplines [psychiatric nurses (n = 5), occupational therapists (n = 2) and social workers (n = 1)].

The knowledge/skills ratings pre and post training are presented in Appendix 2, Figure 3. The mean self-rated knowledge scores increased from pre to post training on all core areas of training (physical health, substance use, running groups and using MI).

In addition, the mean post-training confidence ratings were all high for core components (physical health 81.3%, substance use 79.4%, running groups 86.3%, MI 83.8%) and the mean ratings of the importance of these aspects of support for psychosis were also all high (physical health, 86.9%; substance use, 86.9%; running groups, 87.5%; MI, 92.5%).

Clinicians rated the core training modules on various dimensions, including training quality, resources, level, value, applicability and impact on confidence, as well as on specific skills learnt. Ratings were on a scale from 1 (poor/strongly disagree) to 5 (excellent/strongly agree). The mean ratings and ranges for physical health, substance use, healthy living groups and MI training were 4.4 (4.0–4.9), 4.7 (4.09–5.0), 4.4 (4.0–5.0) and 4.3 (4.0–4.9), respectively.

Qualitatively, clinicians recommended that the primary mode for intervention delivery be individual sessions because of the difficulty in co-ordinating and engaging service users in groups. They also requested that basic CBT be included in the training programme and the manual. They reported that the physical health training was empowering, unleashing potential, and should be more available and that the substance use training should be longer, mandatory and focused more on individual substances. The training on healthy living groups seemed to fine-tune and enhance existing skills, whereas the MI training challenged preconceived ideas and a ‘nursing’ model, and was well shaped for community populations. It gave a structured approach to the intervention, had good links to theory, and provided an opportunity to try things out, but it should be more focused specifically on MI in mental health as opposed to general health.

Phase 3: the Delphi consultation

A total of five expert therapist/researchers from a range of backgrounds (two psychologists, two psychiatrists and one nurse) took part in two rounds of Delphi consultation on the reference book, the therapy manual and the service user handbook. All manuals were redrafted after each round of the process, based on numerical and qualitative feedback, and re-rated such that each underwent three iterations, involving substantial modification of content, format and language. Owing to time pressures in development, no Delphi rating was obtained for the final version. One expert service user/researcher also provided qualitative feedback in both rounds. [See www.amazon.co.uk/Impact-Reference-Improving-Physical-Substance/dp/095688850X (accessed 6 December 2019) for the final versions of the reference guide, the manual and the Better Health Handbook.]

Among those expert therapist/researchers who completed both rounds of the Delphi process, the mean ratings of the therapy manual for user-friendliness, spirit of MI, integration with CBT, usability in the NHS, length and complexity increased over time from 7.3 to 7.6 out of 10, with a rating of 7.4 out of 10 from four participants for the second rating. Similarly, ratings for the reference book increased over time from 6.8 to 7.1 out of 10, with a rating of 7.4 from four participants for the second rating. Finally, ratings for the handbook increased over time from 7.3 to 7.7 but with a lower second rating from four participants in the second iteration of 6.4. This lower rating was as a result of a low mean score from one expert, who also provided extensive feedback that was incorporated in the final handbook.

Two clinicians (one nurse and one occupational therapist) reviewed the therapy manual quantitatively and qualitatively while using it to deliver a ‘test’ session each with a service user (one male and one female). These service users in turn provided qualitative feedback on their experiences of the session. For the therapist who completed both rounds of the Delphi process, the mean rating for the therapy manual increased from 7.3 to 8.2. This related to its usability for day-to-day standard NHS work, including length, complexity and confidence provided.

The expert therapists/researchers generated a large amount of qualitative feedback. The therapy manual was seen as too long, and all manuals needed more simplified language. Specific core recommendations for change included developing the therapy into modules, including practice examples and case studies, providing greater clarity on how to manage interactions with mental health and issues with medication, having built-in guidelines on training and having supervised practice to support manualised therapy delivery. In terms of therapy content and structure, recommendations were made to include a clear definition of each stage of change, what a client might say or do in this stage, and the sessions associated with this stage; that a structure should be followed for each session to include a session number and why and when to do the session, followed by clear aims and objectives and then session content; and that therapists should be advised that they can dip into the sessions and materials relevant to their client and should not be expected to cover everything. This would allow flexibility and avoid overprescription. Explicit links were recommended to be made between the sessions and linked handbook resources. Greater CBT for Psychosis coverage was requested; some concern was raised about offering too brief a CBT for Psychosis section without a preliminary formulation. Another recommendation was to include the application of CBT to enable goals to be reached at the action stage in MI. The various flow charts were often viewed as rather complex. The general point was made that outside IMPaCT therapy in the NHS there would need to be some form of physical health assessment or other means to lead to a health target.

One therapist/researcher rightly suggested that the relatives session may be too short to reduce expressed emotion at home. The relatives session did not aim to be a therapy. At the suggestion of carers who were presented with the IMPaCT study and asked for comments, the session aimed to address some of the misconceptions about and heightened expectations for change that might be held by relatives following the offer of therapy. The session therefore aimed to present only the basic tenets of behaviour change and the notion that most change occurs at a cognitive level before any behavioural change occurs. It aimed to clarify that developing intrinsic motivation will lead to more long-lasting change, as opposed to a forced change that may be short lived.

Clinicians fed back qualitatively that the theory should be clearly separated from the therapy sessions as a separate reference book. The sessions themselves needed to be simplified and clearer in their aims, and more stand-alone as individual session guides, with less ‘jargon’. There was a need to ensure better cross-referencing between the therapy manual sessions and the handbook resources. The MI section was viewed very positively. Overall, echoing the view of the expert therapists/researchers, the consensus was that the initial therapy manual was too long.

The two service users provided only qualitative feedback over the two time points. They confirmed that they found the focus on health to be beneficial, leading to an increase in knowledge, motivation and ideas for improving health, and one reported a significant change in alcohol use (from heavy drinking to 2-week abstinence) following the session. Materials in the handbook were viewed positively. One expert service user/researcher provided specific and detailed feedback on the handbook. Key points included, again, using simpler language; giving greater explanation of how to use and complete the various resources, and what they are; including colour and pictures to liven up the handbook; and including positive language and hopeful messages about change. Consistent with expert therapists, a suggestion was made that terms such as ‘heavy’ or ‘light’ smoking/substance use needed to be defined; in addition, benchmarks and thresholds, for example for problematic waist circumference, should be provided for mainstream NHS use. In the research therapy, this will be provided as part of the IMPaCT assessment.

Study design and setting

A pragmatic, multicentre, two-arm, parallel-cluster, RCT design was used and the HPI was integrated into routine services across five mental health trusts in South London, Kent, Sussex, Somerset and Staffordshire, including both urban and rural populations. The Consolidated Standards of Reporting Trials (CONSORT) cluster trial extension standards guided the planning and implementation of this study. Ethics approval was obtained from the joint South London and Maudsley and the Institute of Psychiatry NHS Ethics Committee (REC reference number 09/HO80/41). Colleagues with lived experience, including both service users and carers, were involved throughout the research, from contributing to funding applications, to managing steering groups, to coauthoring the papers arising from the work.

Participants

Recruitment took place between 1 March 2010 and 1 July 2014. Eligible participants were aged between 18 and 65 years and had an established diagnosis of a psychotic disorder (ICD-10114 diagnosis F20–29, F31.2 or F31.5). Patients were excluded if they (1) had a primary diagnosis of learning disability, (2) had a pre-existing physical health problem that would independently affect the metabolic measures (as judged by medical investigators), (3) were pregnant or < 6 months post partum or (4) had a life-threatening or terminal medical condition. We did not recruit from FEP services. Recruitment started on 1 March 2010, with first randomisation on 9 August 2010.

Study procedure

There were two waves of participant recruitment. First, all permanently employed community care co-ordinators with a minimum of four psychosis patients on their caseload in participating CMHTs were approached in random order (using a random number generator) and invited to participate. Following informed consent from individual care co-ordinators, patients from their caseload who met the inclusion criteria were identified. These patients were then approached and invited to participate, also in a random order, until either four participants consented or all eligible patients had been approached. Baseline measures were then taken from the consenting patients. After baseline assessments of all consenting patients on a care co-ordinator’s caseload were completed, care co-ordinators were randomised and stratified by borough using randomisation blocks of random sizes to deliver IMPaCT therapy or TAU alone to their own current patients (cluster). Both researchers and the statistician remained blind to treatment allocation for the duration of data collection.

In the treatment arm, the intervention (IMPaCT therapy) was provided by a patient’s usual community care co-ordinator. In the first 3 months following a patient’s randomisation, their community care co-ordinator received a 4-day IMPaCT training course. This covered physical health and substance misuse, MI, CBT techniques for running group sessions, and how to deliver IMPaCT therapy and health promotion, and it was given by trainers in MI and in CBT. An evaluation of this is reported in Appendix 3. Throughout the subsequent 9-month intervention, participating care co-ordinators were offered fortnightly supervision in IMPaCT therapy and all care co-ordinators were offered a 1-hour training session in best practice for physical health awareness to ensure more standardised TAU.

The intervention was rolled out into routine care by holding discussions with management, which confirmed their strong support, and then approaching local team management about the practicalities, and then informing the community teams about the RCT and offering the 1-hour education session as above. Further practicalities were addressed at the 4-day training sessions in the intervention group, including advice on planning group sessions if needed. Day-to-day practical issues that arose were dealt with in the fortnightly supervision sessions or, if these were at the team, borough or trust level, the relevant layer of management was involved.

Outcome measures

A change in outcome was defined as a difference from baseline and (1) on completion of the supervised intervention, at 12 months, and (2) 3 months after the end of treatment, at 15 months. Baseline and follow-up assessments were conducted during face-to-face interviews by researchers within established time windows (–6 weeks/+4 weeks at 12 months’ follow-up and ± 4 weeks for 15 months’ follow-up). Data collected outside these times were recorded but used only for sensitivity analyses and not for the main analysis.

Primary outcome

The primary outcomes were the physical and mental health component scores of the SF-36,178 measured at 12 and 15 months. We originally proposed the SF-36 as an integrated score but revised the protocol so that each of the two components was a primary outcome.

Secondary outcomes

Secondary outcome measures were (1) physical health measures (levels of total HDL and low-density lipoprotein cholesterol, triglycerides, HbA_1c and CRP; anthropometric measurements, including waist circumference, BMI, BP; MetS defined according to the International Diabetes Federation criteria);13 (2) substance use measures [alcohol use recorded using the AUDIT,121 smoking prevalence and cigarettes per day using the Nicotine Dependence Questionnaire,120 use of cannabis and other illegal substances (e.g. opiates, methamphetamine, cocaine) using the Time Line Follow Back];72 (3) lifestyle measures (dietary patterns quantified using the DINE179 and physical activity using the short-form International Physical Activity Questionnaire);116,180 and (4) mental health status (using PANSS,115 GAF,117 SF-36 and Montgomery–Åsberg Depression Rating Scales). 181

Sample size

Power analyses were performed for the two subscales measures: physical and mental health components of the SF-36 QoL scale. 178 To detect a clinically significant reduction of 5 points (using 80% power, a 5% alpha level and two-tailed assumption), and to allow for a 20% loss of care co-ordinators and an additional 30% loss of patients to follow-up, a total sample size of 70 care co-ordinators, yielding 280 participants, was needed for the physical scale (d = 0.5), and 98 care co-ordinators, yielding 392 participants, were needed for the mental health scale (d = 0.42).

Statistical analysis

Primary statistical analyses was based on the intention-to-treat principle and aimed to estimate the difference in mean outcomes between participants randomised to HPI and those randomised to TAU at 12 and 15 months using mixed-effects models. Bias as a result of missing follow-up data was assessed by comparing the baseline characteristics of those with and those without complete data. In two linear mixed-effects models, the physical and mental health component scores at 12 months and 15 months constituted the dependent variable. ‘Treatment randomisation group’, ‘time’ (with two levels, i.e. 12 and 15 months post randomisation), and the interaction between ‘treatment group and time’, ‘centre’ and the ‘baseline values of physical and mental health component scores’ constitute the fixed part of the model. An unstructured covariance pattern model was used to model the dependency of the repeated observations of the same subject and to account for the dependency of the subjects within a cluster, and the care co-ordinator was included as a random factor. Model assumptions were assessed by visual inspection of the residuals. Standardised effect sizes, using pre-randomisation variability for standardisation, were also reported.

Secondary outcomes were analysed using the same methods as for the primary outcome. For all models, the interaction between treatment group and time was not significant and was removed from the final analyses. Treatment effects are, therefore, estimates for both time points. Logistic mixed models were used for binary outcomes (e.g. smoking) and Poisson mixed models were used for count data (e.g. number of cigarettes per day). Because of the large number of tests, significant results need to be treated as explorative.

Sensitivity analyses

Approximately 15% of the observations were collected outside the time window, so analyses using all available data were repeated as sensitivity analyses.

Handling of missing data

Models were rerun with predictors related to outcome missingness included as further covariates in the model. For the main outcomes, a second sensitivity analysis of missing outcome data, using multiple imputations by chained regression equations, was performed using all available clinical and demographic scores. This was done separately for each treatment group. A further sensitivity analysis was carried out to explore the impact of not missing at random patterns by adding positive or negative values to the imputed data sets.

Complier average effect analyses

In addition to the standard intention-to-treat (efficiency) analysis, we estimate the measure of the treatment impact for compliers only (treatment efficacy) using an instrumental variable approach with randomisation indicator as an instrumental variable. 182 Analyses were carried out using Stata’s^® (StataCorp LP, College Station, TX, USA) ivregress package, with centre, treatment arm and baseline values of the outcome constituting the fixed part of the model and with cluster robust standard errors (SEs) to control for clustering effects of care co-ordinator.

Key findings

A total of 104 care co-ordinators were recruited; 52 (with 213 patients) were randomised to deliver IMPaCT therapy and 52 (with 193 patients) were randomised to TAU. Of the 406 patients, 318 (78%) and 301 (74%) attended the 12- and 15-month follow-ups, respectively. The total sample size was reduced to 263 (64.8%) for 12 months (TAU, n = 132; HPI, n = 131) and to 238 (58.6%) for 15 months (TAU, n = 114; HPI, n = 124) because some of the patients who attended follow-up assessments were not seen within the required time frame. The required sample size based on the power analyses was achieved at both time points.

IMPaCT therapy showed no significant effect on the physical or mental health components of the SF-36 scores compared with TAU at 12 or 15 months. 176 Sensitivity analyses, including missing data analyses, did not alter any conclusion176 (see Appendix 5). A complier average casual effect analysis did not provide evidence of the efficacy of the intervention at 12 months for the physical health component (PHC) (mean difference –2.01, 95% CI –4.18 to 0.16; p = 0.07) or mental health component (MHC) (mean difference 0.02, 95% CI –2.79 to 2.83; p = 0.99).

No effect was observed for cardiovascular risk indicators, except for HDL cholesterol, which showed greater improvement following IMPaCT therapy than following TAU (treatment effect 0.09, 95% CI 0.007 to 0.16; p = 0.034). The 22% of patients who received > 180 minutes of IMPaCT therapy in addition to usual care achieved a greater reduction in waist circumference (–4.2 cm, 95% CI –7.2 to –1.2 cm; p = 0.006) than did control patients. This reduction is clinically significant. There was no difference in the rates of serious adverse events between the groups. The fidelity assessment is detailed in Appendix 3.

Strengths of the trial

The main strength of this trial is that it was a pragmatic study: the design of the HPI was centred on creating an intervention accessible to the greatest possible number of people with psychosis receiving care in CMHTs. This is a traditionally hard-to-reach group; many people are unable or unwilling to attend standard group interventions, which is particularly problematic because of the high levels of cardiometabolic risk. 183

This study’s integrated personalised approach sought to maximise representativeness of the population studied. It addressed both lifestyle choices and substance use and avoided a piecemeal approach to behavioural change. By integrating IMPaCT therapy into established mental health care and by providing the care through a patient’s usual therapist, care could be adapted to the patient’s individual needs. To date, few large-scale, long-term RCTs have attempted to improve health in its widest sense in people with psychosis, instead focusing on a single target and often introducing new therapists to run the trial intervention.

The study recruited to target and was sufficiently powered to detect a difference between the two groups of at least 5 points in physical and mental health component scores (Cohen’s d = 0.42), had good follow-up rates and recruited a diverse, multiethnic sample of people with psychosis.

Limitations of the trial

Continuity of care was affected by multiple reorganisations of staff among the hospital trusts. Staff turnover sometimes delayed recruitment to the trial, as an inclusion criterion was that the staff members expected to be in post for the forthcoming year. Despite this, although not statistically different, staff turnover was slightly greater in the treatment arm than in the control group. However, when patients remained with the same trained care co-ordinator throughout the trial, there were indications that the intervention may result in significantly lower cholesterol levels and more exercise in the treatment group than in the control group. It is known that diminished continuity of care is associated with worse clinical outcomes. 184

Care co-ordinators struggled to deliver six or more 30-minute sessions of the HPI in addition to routine care, despite the training and ongoing supervision, and had limited fidelity to the MI model. This is in keeping with the recognised challenges of delivering targeted psychosocial interventions to people with psychosis that are faced by care co-ordinators in a busy secondary care environment. 185 It is also consistent with the findings of the Health Improvement Profile (HIP) RCT, which trained community mental health nurses to administer health checks to their patients and found that the volunteer nurses administered health checks to only a minority of participating patients. 186 However, given that the subset who received > 180 minutes of additional time devoted to IMPaCT therapy had a clinically significant reduction of –4.2 cm (95% CI –7.2 to –1.2 cm) in their waist circumference, a focus on implementation strategies in future work may be important.

A pilot trial was not conducted because the intervention was built on two pre-existing, successful, published interventions in this population, used a recognised behaviour change theory and incorporated additional recommendations for improving the preceding interventions. However, the longer combined intervention and the delivery model using non-expert care co-ordinators in a UK setting were not tested previously and raised logistical issues, which perhaps could have been identified and addressed more effectively in a large-scale pilot trial, although it is doubtful that they would have been identified in a small pilot.

The HPI was broad and participants self-selected, the target behaviours to focus on. It is conceivable that a more structured intervention targeting specific health behaviours, such as those described elsewhere,187,188 could be more successful in demonstrating statistically significant improvements in circumscribed aspects of the physical health of this population. Nonetheless, a longer-term, sustainable and integrated approach to overall health behaviours and health outcomes is urgently needed to allow successful and accessible routine care in the NHS.

Data collection

An adapted version of the Client Service Receipt Inventory (CSRI)189 was used to measure individual-level resource use. It covered the use of (all-cause) secondary and community-based health and social care services, prescription medication, time off work, and social security benefits received by participants and carers. It was administered as a retrospective, self-report, questionnaire-based interview conducted by assessors who were blind to treatment allocation. It covered the previous 6-month period at baseline and at the 12-month follow-up, and the previous 3-month period at the 15-month follow-up.

Health-related QoL was assessed with self-report questionnaires at baseline and at 12 and 15 months, and the SF-36 and EuroQol-5 Dimensions, three-level version (EQ-5D-3L),190 were used to estimate QALY gains.

Costs

Unit costs were applied to individual-level resource use data to calculate the total costs per participant from two cost perspectives: health and social care, and societal. All-cause health and social care costs included the trial interventions (HPI for the intervention group and standard information session for the control group), specialist accommodation, inpatient services, outpatient services, community-based services, community-based professionals and prescription medications. Societal costs included the same health and social care costs plus third-sector (charity) costs, lost productivity costs due to absence from work (if in employment during the assessment period), participant out-of-pocket expenditure on community-based day services and selected social security benefits. Lost productivity costs were capped at 5 days per week (maximum of 130 days for 6-month data and 65 days for 3-month data).

Unit costs are detailed for reference in Appendix 4, Table 9. In brief, the unit costs for most hospital and primary care services were obtained from the NHS Reference Costs 2010–11191 (inflated to 2011–12 prices using the Hospital and Community Health Services Pay and Prices Index or retail price index as appropriate) and the Unit Costs of Health and Social Care 2012 compilation. 192 Medication unit costs, taken from the British National Formulary,193 were converted into cost per milligram based on the most cost-efficient pack size, choosing maintenance doses over initial treatment doses and generic formulations over branded ones to obtain conservative estimates.

All costs are reported in Great British pounds (GBP) at 2011–12 prices. Discounting was applied to cost and outcome data related to the 12- to 15-month assessment period. Intervention costs were not discounted before being added to the 12- to 15-month costs because they were incurred in the first year. Discounting was applied at a rate of 3.5%. 194

Cost of the health promotion intervention

We included the following resource components to estimate the costs associated with the intervention:

• the production of manuals (excluding the development work)

• the training of care co-ordinators

• the ongoing supervision of care co-ordinators

• the implementation of the intervention by care co-ordinators with trial participants.

The resources and costs associated with each of these components are available for reference in Appendix 4, Table 9. The costliest of the four components was the implementation with patients, at a mean cost of £92.31 per case. To estimate the total participant-level intervention costs, we first averaged out general costs using a ‘top-down’ approach (i.e. we summed the costs of the manual, training and supervision, and divided this by the recommended number of patients under the care of each care co-ordinator). To this average we added the individual-level implementation costs.

Health promotion intervention manual costs

Care co-ordinators randomised to the intervention arm were provided with a copy of the IMPaCT manual, the IMPaCT reference guide and the Better Health Handbook. The time inputs to the intellectual development of these documents were considered a sunk cost and were, therefore, excluded. However, we included the reproduction costs.

Health promotion intervention training costs

Costs related to training care co-ordinators included the following components: the cost of trainer time (contact and non-contact), the cost of care co-ordinator attendance time and the cost of materials necessary to deliver the training courses. The training programme was delivered over 4 days by members of the research team in a standardised manner, with specific staff dedicated to implementing certain parts of the programme. The total number of hours and the number of sessions for each staff type were calculated as detailed in Appendix 4, Table 10. Preparation time was estimated by multiplying the number of training sessions by an assumed preparation time per session of 10 minutes. Costs were calculated by multiplying the appropriate unit costs by the number of minutes for each staff type.

Each of the care co-ordinator attendees was an NHS band 6 nurse. Each training programme took 32 hours including breaks (09.00 to 17.00 for 4 days). Costs were calculated by multiplying the appropriate unit costs by this time.

Materials were broken down into one-off costs and repeat costs. One-off costs included BP machine, weighing scales, tape measure, flip-chart stand and memory stick. Repeat costs included flip-chart paper, flip-chart pens, paper and pens.

Health promotion intervention supervision costs

At the outset of the study, it was envisaged that ongoing supervision of care co-ordinators would take place in a combination of individual and group sessions, led by a band 6 nurse trainer and a band 8 nurse trainer, for 9 months after attendance at the training programme. However, because of practical limitations, group supervision was unable to be provided, so all supervision took place at the individual level in 90-minute sessions. Costs were calculated by multiplying appropriate unit costs by the number of minutes spent in supervision by both the supervisors and the supervisees.

Health promotion intervention participant implementation costs

Similarly, it was originally envisaged that the intervention would be delivered in a combination of individual and group sessions led by care co-ordinators. However, group intervention sessions were not provided, largely because of practical limitations, so all implementation occurred in individual sessions. To estimate the costs of the intervention at the level of individual patient participants, we multiplied the cost of each individual therapy session by the care co-ordinator estimates of the proportion of time in each session spent delivering an aspect of the HPI.

Cost of the control intervention

Care co-ordinators in the control group of the trial were given a one-off information-giving session on mental and physical health issues. This involved a consultant psychiatrist delivering the session over a 1.5-hour period. The appropriate unit costs were applied and the cost per care co-ordinator was then divided by the recommended number of patients under the care of each care co-ordinator.

Outcome measures

Cost-effectiveness analyses were based on the joint primary outcome measures: the SF-36 MCS score and SF-36 PCS score. Cost–utility analyses were based on QALYs derived from the EQ-5D-3L and the SF-36 (US version 1) via the Short Form questionnaire-6 Dimensions (SF-6D). Appropriate utility weights were attached to health states for each measure at baseline and at 12 and 15 months. 178,190 QALY gains between 12 months and 15 months were then calculated using the total area under the curve approach with linear interpolation between assessment points. 195

Analyses, missing data and sensitivity analyses

For the economic evaluation analysis, see Appendix 4.

Key findings

The economic evaluation showed no evidence of a clear difference in health and social care costs between the two trial groups. Societal costs may be higher in the intervention group. Alongside a lack of additional benefit, there is no evidence that the intervention is cost-effective. The integrated mental and physical health-care intervention itself involved no significant additional resource use or associated costs. However, it is unclear to what extent this finding reflects a true lack of impact on costs and outcomes, or whether or not it reflects the fact that, because very few participants in the intervention group received the intervention as planned, both groups ultimately received similar care.

Limitations

The economic evaluation had some limitations. Data on resource use and, therefore, on costs were collected using the self-reported CSRI. This makes the data subject to participant recall bias. However, there is evidence for the reliability of self-reported resource use data in similar populations196,197 and we considered the approach necessary given the multisite and multiperspective nature of this study; the lack of integration of all relevant health and social care sector client records necessitated such an approach to avoid restriction to a narrow definition of health and social care. Furthermore, at least some resource use would still have been needed to be collected via self-report (i.e. welfare benefits and employment losses) to enable a societal perspective to be considered, which we considered of particular relevance for a patient group whose health and care needs can have economic impacts on multiple sectors of society. Additionally, there is no reason to believe that any biases related to data collection would be unbalanced between the two trial groups, particularly because the CSRI was administered by blinded assessors.

A further limitation is that we may have double counted resource use associated with the HPI. We collected this information separately from care co-ordinators rather than from patient participants to avoid unblinding the assessors conducting the participant interviews. However, this carried the risk of participants reporting resource use related to the intervention in response to questions about their service use. Although this may skew absolute estimates of costs for the intervention arm, any bias in the calculation of treatment effects would work against the intervention group by overestimating the intervention cost.

There has been some discussion around the validity of using the SF-36 and EQ-5D-3L with study participants who have mental health problems, especially those with schizophrenia and other psychoses. 198 Brazier et al. 198 suggest that neither scale performs particularly well in terms of quantitative testing against psychometric criteria and that both have a limited coverage of domains identified as relevant by people with mental health problems. Thus, it is unclear whether or not the lack of QALY difference between the two trial groups reflects a lack of intervention effect or limitations associated with the measurement properties of these two health-related QoL measures. However, given the lack of effect based on the SF-36 MCS and PCS, and all other outcome measures, it is unlikely that there was a difference in QALYs that we have been unable to detect.

Finally, the time horizon of the evaluation is likely to have been insufficient to identify all relevant outcomes for this patient group, particularly given the longer-term impacts of physical health problems. However, it is unlikely that any effects of the intervention would transpire in the longer term if absent in the short term.

Work package 1: PUMP

Reis-Marques T, Smith S, Bonaccorso S, Gaughran F, Kolliakou A, Dazzan P, et al. Sexual dysfunction in people with prodromal or first-episode psychosis. Br J Psychiatry 2012;201:131–6.

Crews M, Lally J, Gardner-Sood P, Howes O, Smith S, Murray RM, et al. Vitamin D deficiency in first episode psychosis: case–control study. Schizophr Res 2013;150:533–7.

O’Brien C, Gardner-Sood P, Corlett SK, Ismail K, Smith S, Atakan Z, et al. Provision of health promotion programmes to people with serious mental illness: a mapping exercise of four South London boroughs. J Psychiatr Ment Health Nurs 2014;21:121–7.

Kolliakou A, Castle D, Sallis H, Joseph C, O’Connor J, Wiffen B, et al. Reasons for cannabis use in first-episode psychosis: does strength of endorsement change over 12 months? Eur Psychiatry 2015;30:152–9.

Russell A, Ciufolini S, Gardner-Sood P, Taylor H, Bonaccorso S, Gaughran F, et al. Inflammation and metabolic changes in first episode psychosis: preliminary results from a longitudinal study. Brain Behav Immun 2015;49:25–9.

Colizzi M, Carra E, Fraietta S, Lally J, Quattrone D, Bonaccorso S, et al. Substance use, medication adherence and outcome one year following a first episode of psychosis. Schizophr Res 2016;170:311–17.

Lally J, Ajnakina O, Stubbs B, Williams HR, Colizzi M, Carra E, et al. Hyperprolactinaemia in first episode psychosis – a longitudinal assessment. Schizophr Res 2017;189:117–25.

Theleritis C, Bonaccorso S, Habib N, Stahl D, Gaughran F, Vitoratou S, et al. Sexual dysfunction and central obesity in patients with first episode psychosis. Eur Psychiatry 2017;42:1–7.

Gaughran F, Stahl D, Stringer D, Hopkins D, Atakan Z, Greenwood K, et al. Effect of lifestyle, medication and ethnicity on cardiometabolic risk in the year following the first episode of psychosis: prospective cohort study. Br J Psychiatry 2019;215:712–19.

Work package 3: IMPaCT randomised controlled trial

Gaughran F, Stahl D, Ismail K, Atakan Z, Lally J, Gardner-Sood P, et al. Improving physical health and reducing substance use in psychosis – randomised control trial (IMPaCT RCT): study protocol for a cluster randomised control trial. BMC Psychiatry 2013;13:263.

Gardner-Sood P, Lally J, Smith S, Atakan Z, Ismail K, Greenwood KE, et al. Cardiovascular risk factors and metabolic syndrome in people with established psychotic illnesses: baseline data from the IMPaCT RCT study. Psychol Med 2015;45:2619–29.

Stubbs B, Gardner-Sood P, Smith S, Ismail K, Greenwood K, Farmer R, Gaughran F. Sedentary behaviour is associated with elevated C-reactive protein levels in people with psychosis. Schizophr Res 2015;168:461–4.

Stubbs B, Gardner-Sood P, Smith S, Ismail K, Greenwood K, Patel A, Farmer R, Gaughran F. Pain is independently associated with reduced health related quality of life in people with psychosis. Psychiatry Res 2015;230:585–91.

Lally J, Gardner-Sood P, Firdosi M, Iyegbe C, Stubbs B, Greenwood K, et al. Clinical correlates of vitamin D deficiency in established psychosis. BMC Psychiatry 2016;16:76.

Gaughran F, Stahl D, Ismail K, Greenwood K, Atakan Z, Gardner-Sood P, et al. Randomised control trial of the effectiveness of an integrated psychosocial health promotion intervention aimed at improving health and reducing substance use in established psychosis (IMPaCT). BMC Psychiatry 2017;17:413.

Heslin M, Patel A, Stahl D, Gardner-Sood P, Mushore M, Smith S, et al. Randomised controlled trial to improve health and reduce substance use in established psychosis (IMPaCT): cost-effectiveness of integrated psychosocial health promotion. BMC Psychiatry 2017;17:407.

Harrison RNS, Gaughran F, Murray RM, Lee SH, Cano JP, Dempster D, et al. Development of multivariable models to predict change in body mass index within a clinical trial population of psychotic individuals. Sci Rep 2017;7:14738.

Onwumere J, Shiers D, Gaughran G. Physical health problems in psychosis: is it time to consider the views of family carers? Front Psychiatry 2018;9:668.

Stubbs B, Vancampfort D, Hallgren M, Firth J, Veronese N, Solmi M, et al. EPA guidance on physical activity as a treatment for severe mental illness: a meta-review of the evidence and position statement from the European Psychiatric Association (EPA), supported by the International Organisation of Physical Therapists in Mental Health (IOPTMH). Eur Psychiatry 2018;54:124–44.

Westman J, Eberhard J, Gaughran FP, Lundin L, Stenmark R, Edman G, et al. Outcome of a psychosocial health promotion intervention aimed at improving physical health and reducing alcohol use in patients with schizophrenia and psychotic disorders (MINT). Schizophr Res 2019;208:138–44.

Measures

At baseline, after giving informed, written consent, each participant was asked to complete a battery of measures. This included sociodemographics, clinical measures and health-related QoL, measured using the EQ-5D-3L. 190 Clinical measures relevant to these analyses were clinical indicators of MetS/cardiometabolic risk (obesity, triglycerides, HDL cholesterol, BP and fasting plasma glucose). 13 The EQ-5D-3L was again administered at the 12-month follow-up to enable the calculation of 1-year QALY gains.

Data on mental health care hospital admissions for the 12 months prior and 12 months following presentation for FEP were manually extracted from electronic patient record systems in each of the relevant NHS trusts using a pro forma specifically designed for the task to standardise the methodology and information that was collected. Core data included admission and discharge dates of each admission to enable the calculation of length of stay per admission, and ward type to enable more accurate estimation of costs.

Unit costs

Unit costs were applied to individual-level inpatient days according to ward type to calculate total mental health hospital admission costs per participant over 1 year. Unit costs were obtained from NHS Reference Costs for the year 2010–11 and inflated to 2011–12 prices using the Hospital and Community Health Services Pay and Prices Index. 192 We used the Reference Costs 2010–11191 because they offered more detailed cost data according to the type of mental health service, which is not offered in the 2011–12 Reference Costs. Unit costs are detailed in Table 1. All costs are reported in GBP at 2011–12 prices. Discounting was not necessary as all costs related to a 1-year period.

Analyses

For the purpose of univariate associations with costs and QALYs, individual metabolic factors were examined in both binary (present or not) and continuous form. Furthermore, the total number of metabolic factors present was investigated.

To estimate QALYs, we applied general population utility weights to EQ-5D-3L health state measurements at baseline and at 12 months. 124 QALY gains between baseline and 12 months were then calculated using the total area under the curve approach with linear interpolation between assessment points. 195

Data were analysed using Stata 11. Despite the skewed nature of the cost data, the mean and SDs for cost data are reported as recommended. 224 Associations with costs and QALYs were explored using the stepped approach. 225 Univariate associations of metabolic and sociodemographic variables with costs were examined using non-parametric bootstrap regressions (ordinary least squares); we conducted each regression both with and without a covariate for baseline costs. All independent variables that were associated (p < 0.1) with costs in these univariate analyses were then carried forward for inclusion in a multivariate non-parametric bootstrap regression. We examined the results of the multivariate regression to identify any variables that no longer remained significantly associated with costs (p > 0.1) and then reran the multivariate regression without those variables. Variables that did not appear significantly associated with costs in the univariate regressions were then added into the model and retained if they significantly improved the model or rejected if they did not. The same process was repeated for the analysis of associations with QALYs.

Sample characteristics

Of the 293 participants, 190 (65%) were male and 136 (46.4%) were of white ethnic origin. The mean age was 30.6 years (SD 10.5 years). Seventy-six (31.8%) lived alone and 155 (66.2%) were unemployed.

Data completeness

A total of 274 out of 293 participants (93.5%) had full data on resource use/costs for both the 12 months prior to and the 12 months following presentation for FEP, which were gathered from electronic health records. However, fewer people had all relevant data on metabolic factors and QALYs. This resulted in 100 participants with all relevant data for the examination of associations with admission costs and 55 participants with all relevant data for the examination of associations with QALYs. Table 2 shows that the two subgroups available for analyses were similar to the full sample in terms of a range of baseline characteristics. However, both subgroups had lower admission costs at baseline than the full sample. The potential implications of missing data are explored in Discussion.

Resource use and costs

Among the 293 participants, we were able to extract data for 278 on the number of inpatient days for the year prior to FEP. Of these, 243 participants had no inpatient days prior to FEP. Of the 35 participants who had had a mental health admission, this was mostly because of a few (1–5) days admitted for symptoms during the prodromal period before a diagnosis of psychosis was assigned. Four participants had total admission days ranging between 8 and 172 days and the most likely reason for this was an evolving clinical picture before the diagnosis of psychosis became clear. The mean cost of admissions in the year prior to FEP was £443 (SD £5814, range £0–95,957).

We were able to extract data on the number of inpatient days for the year following presentation for FEP for all 279 participants. Twenty-eight participants had no admissions. The overall mean total was 63 days [SD 69 days, range 0–367 days (exceeds 365 as a result of accounting for movement between wards)]. The associated mean cost was £25,376 (SD £34,411, range £0–236,945).

The proportion of participants who had at least one admission and the mean number of inpatient days for the year prior to FEP were similar in the full sample and in the subsample who had all required data for the analysis of admission costs (i.e. they also had full metabolic and sociodemographic data) (Table 3). This was also true for the 1 year following presentation for FEP. However, the mean cost of admissions in the year prior to FEP was lower for the subsample with full data than for the full sample [£77 (SD £276) vs. £433 (SD £5814)].

Metabolic syndrome factors and admission costs

Univariate analyses suggested that being male, obese or of non-white ethnicity was associated with higher admission costs during the 1-year follow-up (p < 0.1), whereas being older or having reduced HDL cholesterol (< 1.03 mmol/l for men and < 1.29 mmol/l for women) was associated with lower costs during the 1-year follow-up (p < 0.1) (Table 4). When these factors were carried through to a multivariate analysis, all except ethnicity remained significantly associated with costs (Table 5).

Health state utility

Of the 293 participants in the full sample, 89 had full EQ-5D-3L data at both baseline and follow-up to enable the calculation of QALY gains for the year following presentation for FEP. The mean QALY gain for these 89 participants was 0.84 (SD 0.17). The subsample was further reduced to 55 when accounting for availability of relevant metabolic and sociodemographic data; however, this subsample appeared broadly representative of the full sample of participants, who had a comparable mean QALY gain of 0.86 (SD 0.14).

Metabolic syndrome factors and quality-adjusted life-years

Both unadjusted and adjusted (for baseline utility value) univariate analyses suggested that, at baseline, having a MetS, being obese, having reduced HDL cholesterol (binary variable), having higher HDL cholesterol (continuous variable) and having more metabolic indicators were associated with lower QALY gains (Table 6). Having raised fasting glucose or higher triglycerides (adjusted coefficient –0.03, 95% CI –0.07 to –0.00) was associated with lower QALY gains in adjusted analyses only.

When examining these seven variables in a multivariate analysis adjusted for baseline utility, only HDL level (as a continuous variable) remained statistically associated with QALYs, with higher HDL cholesterol associated with greater QALY gains (HDL 0.09, 95% CI 0.02 to 0.17; p = 0.013) (Table 7).

Resource use

Resource use pattern data suggest that both arms were broadly balanced in their use of core services both before and during the study. These data are available for reference in Tables 18–20. As would be expected for this group of patients, service use is very broad in both nature and sector, illustrating the complexity of their care provision. These data were not compared statistically because the economic evaluation was focused on costs and cost-effectiveness/utility, and to avoid problems associated with multiple testing.

Costs

There were 52 care co-ordinators in the intervention group and a total of 213 patient participants with whom they implemented the HPI. The mean number of trial participants per care co-ordinator was four (range 1–10 participants). The total average cost per case in the intervention arm was £226.40, compared with £3.52 in the control arm.

Health and social care and lost productivity formed the largest components of total societal costs. Costs at baseline, 12 months and 15 months are summarised for reference in Table 21. The mean costs of all summary components were similar between trial arms at all assessment points, except the cost of the intervention, which was naturally higher in the intervention group given the additional inputs required compared with the control group (adjusted mean difference £311, 95% CI £267 to £355), and costs borne by charities, which were higher in the intervention group at 12 months (adjusted mean difference £80, 95% CI £9 to £151).

Comparisons of the total costs from both health and social care and societal perspectives at 15 months suggested no difference between the trial arms, although the 95% CIs suggested a tendency for societal costs to be higher in the intervention arm (Table 22). Imputing the missing total costs at 15 months for the sensitivity analyses confirmed this (Table 23).

Outcomes

There were no differences in any outcome at any time point (Table 24). As with cost data, imputing missing outcome data for those lost to follow-up (Table 25) did not alter the conclusions drawn from the base-case analyses of only those with available data.

Cost-effectiveness and cost–utility analyses

The two trial arms showed the same costs and same outcomes for all prespecified 15-month cost–outcome combinations using the base-case approach (Table 26); therefore, it was not necessary to compute any ICERs. Furthermore, none of the four sensitivity analyses altered the conclusions about between-group differences in total costs or outcomes.

The cost-effectiveness planes from health/social care and societal perspectives for all four outcomes are in Figures 4–11, showing no difference between the groups in costs or outcomes. Although cost-effectiveness planes for EQ-5D-3L-based QALYs from the health and social perspective indicate a tendency for estimates to predominate in the north-east or south-east quadrants (indicating higher or lower costs and better outcomes), those for the SF-36 MCS and SF-36 PCS show more estimates falling into the north-west or south-west quadrants (indicating higher or lower costs and worse outcomes). Cost-effectiveness planes from the societal perspective (see Figures 8–11) all suggest that mean costs in the intervention arm are likely to be higher than in the control arm, as was suggested by the 95% CIs for estimates of mean cost differences (see Tables 22 and 23).

FIGURE 4.

Cost-effectiveness plane of mean differences in EQ-5D-3L-based QALYs and health and social care costs.

FIGURE 5.

Cost-effectiveness plane of mean differences in SF-36-based QALYs and health and social care costs.

FIGURE 6.

Cost-effectiveness plane of mean differences in SF-36 MCS and health and social care costs.

FIGURE 7.

Cost-effectiveness plane of mean differences in SF-36 PCS and health and social care costs.

FIGURE 8.

Cost-effectiveness plane of mean differences in EQ-5D-3L-based QALYs and societal costs.

FIGURE 9.

Cost-effectiveness plane of mean differences in SF-36-based QALYs and societal costs.

FIGURE 10.

Cost-effectiveness plane of mean differences in SF-36 MCS and societal costs.

FIGURE 11.

Cost-effectiveness plane of mean differences in SF-36 PCS and societal costs.

From a health and social care perspective, the probability that the HPI is cost-effective does not exceed 0.4 for any of the examined willingness-to-pay thresholds for QALY gains (based on either the SF-36 or the EQ-5D-3L). When considering willingness to pay for point improvements in the PCS and MCS of the SF-36, the probability of the intervention being cost-effective starts at 0.34 at a willingness to pay of £0 but reduces to around 0.2 for willingness-to-pay values between £5000 and £50,000 (Figures 12–15).

FIGURE 12.

Cost-effectiveness acceptability curves for SF-36- and EQ-5D-3L-based QALYs from a health and social care perspective.

FIGURE 13.

Cost-effectiveness acceptability curves for the PCS and MCS of the SF-36 from a health and social care perspective.

FIGURE 14.

Cost-effectiveness acceptability curves for SF-36- and EQ-5D-3L-based QALYs from a societal perspective.

FIGURE 15.

Cost-effectiveness acceptability curves for the PCS and MCS of the SF-36 from a societal perspective.

Conclusions are similar from a societal perspective: the probability that the HPI is cost-effective does not exceed 0.12 for any of the willingness-to-pay thresholds examined for QALY gains (calculated from either the SF-36 or the EQ-5D-3L), or 0.21 for the thresholds examined for the point improvements in the two SF-36 component scores.