Project objectives

Aim:  to understand the volume and content of general practice workload, and how these relate to characteristics of general practices, the medical workforce, and different types of geographical area.

Objectives:

1. To establish the feasibility of conducting a national morbidity study using data routinely collected in GP computerised records
2. To establish the feasibility of developing linkages between CPRD data and other data sets (HES, QOF, HSCIC, ONS, GPPS)
3. To examine the pattern of disease presenting to English general practice, and how this relates to the socio-demographic characteristics of the population and characteristics of the clinician (e.g. GP or nurse).
4. To compare the results with previous morbidity studies
5. To explore relationships between practice consultation rates and hospital admission rates, and whether this relationship is moderated by characteristics of practices, their staffing profile, practice organisation (e.g. ratio of phone vs face to face consultations; or GP vs nurse consultations) and/or by socio-demographic factors.
6. To explore relationships between GP workload and outcomes (QOF performance, patient satisfaction, mortality) and whether any such relationship is moderated by characteristics of practices, their staffing profile, or socio-demographic factors.

Design:  Cross-sectional analysis of patient-level consultation data for a complete year from the Clinical Practice Research Datalink (CPRD), with linkages to Hospital Episode Statistics, Mortality Statistics, QOF targets, patient satisfaction data and practice-level staffing information.

Changes to the project originally outlined in the proposal

The project has taken longer than anticipated due to the complexities of categorising the codes used in routine care and relating these to clinical outcomes, diagnoses or reasons for consultation.  We had planned to use mapping software produced by the HSCIC to map Read codes to ICD10 codes, but much of the clinical content of consultations could not be ascribed to diagnostic codes.  We therefore explored other mechanisms to describe the content of consultations and, by using the International Classification of Primary Care coding systems (ICPC-2), we achieved complete categorisation of all codes recorded in the consultation file.

We had originally planned to describe patterns of consulting behaviours in terms of 4 primary outcomes:  consulting prevalence attributed to a condition; incidence; consultation rates and proportion of consultations including content regarding each condition.  However, without recourse to the prior medical record it was not possible to define incidence accurately, and therefore we have not included this outcome. 

All other objectives have been achieved.

Brief summary

Sampling frame:

Clinical consultations with GPs and nurses occurring in all practices included in the Clinical Practice Research Datalink (CPRD) during a one year period April 2013- March 2014 were analysed.    For objectives 1 – 4, the study population was a 10% age and sex stratified random sample of all acceptable¹¹ patients registered for some or all of the year of analysis.  For objectives 5 and 6, the complete consultation data were utilised. Person years of observation were calculated for each age and gender strata.

Objective 1:

In order to establish the feasibility of conducting a national morbidity study using data routinely collected in computerised GP records, we explored several approaches to categorising the MEDCODE codes within the medical records. Firstly, we obtained a cross

mapping schema that links ICD codes to READ codes and examined the possibility of using that mapping process to categorise the reasons for consultations.  However, only 15% of the codes used in the CPRD data could be linked, or matched, in the cross-map.  This is likely to be due to coded entries in primary care containing information about processes and procedures, symptoms, signs and clinical concerns rather than confirmed diagnoses.  Next, we examined the top 300 mostly frequently used codes that were unmatched to ICD codes, and manually mapped them.  This resulted in 66% being successfully identified.  Simultaneously we examined the proportion of consultations that included READ codes from Chapters A to Z or 1 (diagnostic or symptom codes).  Approximately 42% of consultations contained codes that could be mapped to READ codes – again indicating that a large proportion of primary care workload is related to issues that are not formally given a diagnostic code.  The third approach was to utilise and refine a set of lists developed by our collaborators in Cambridge that replicated the code lists that were used to estimate multi-morbidity in primary care in Scotland.  However, these codes only included 36 chronic conditions and therefore did not cover acute conditions, again failing to provide sufficient coverage of the codes used in practice.  The final, and most successful approach has been to map all the codes used within the year of analysis to the International Coding in Primary Care (ICPC-2) scheme.  The major advantage of this system is that it includes symptoms and concerns in addition to diagnoses. This was a manual task where each and every code was translated by a data analyst and a GP and then checked by a second GP.  Discrepancies were resolved by discussion.

We therefore conclude that after a lot of alternative approaches, we have identified a mechanism to describe and catalogue the content of consultations.  Extending this process to other years of data would necessitate the manual mapping process to be conducted on any further codes that had been utilised outside of our year of analysis. 

Objective 2:

To establish the feasibility of developing linkages between CPRD data and other data sets (HES, QOF, HSCIC, ONS, GPPS):  Linkages to HES and ONS are routinely available, so this did not pose any problems.  QOF reported data, patient satisfaction with the GP service (from the General Practice Patient Survey) and practice staffing levels (from HSCIC) are all available via the relevant websites.  The challenge for us was then to link these identifiable publicly available data with the anonymised data from CPRD whilst avoiding any deductive identification of CPRD practices.  Therefore, we converted individual practice-based scores, for QOF indicators and Patient Satisfaction Scores, into performance deciles, only included the minimum number of variables from these datasets to answer the research questions, and provided this data with the published practice identifiers to CPRD.  CPRD then removed practices that are not included in the CPRD-linked data, aggregated data where necessary to ensure there were no cross-tabulations containing less than 5 practices, attached the CPRD practice code attached and removed practice identifiers, before returning the anonymised data to us.

Objective 3:

To examine the pattern of disease presenting to English general practice, and how this relates to the socio-demographic characteristics of the population and characteristics of the clinician (e.g. GP or nurse).

Three main outcomes have been calculated:  the consulting prevalence (defined as the number of patients who consulted at least once during the study year with a given condition divided by the person years of observation);  the condition specific consultation rate/person years (the number of consultations in a year regarding the condition divided by the person 

years registered in practice); and the proportion of consultations containing issues relevant to the conditions (number of consultations in a year with a given condition divided by the total number of consultations).  These have all been reported by age and gender specific strata for each of the 17 ICPC-2 chapters.  Heat maps for each outcome have been produced to visually describe consultation based on deciles within each reporting metric.

We are preparing these data for publication in a high impact journal.

Objective 4:

To compare the results with previous morbidity studies.

We have examined some exemplar conditions in order to compare our results with previously conducted morbidity studies.  Exemplar conditions were chosen as either conditions that are included in the Quality and Outcomes Framework (QOF) and therefore will be used to estimate the face validity of our categorisation (asthma, cancer, depression, diabetes and hypertension), or the prevalence or workload associated with that condition may have altered over time (excema, cancer, depression, rheumatoid arthritis, and shoulder pain)We are reporting these comparisons using a narrative commentary, recognising that some of the differences observed are likely to be due to differences in record keeping and coding. Some differences relate to changes in consulting behaviour, whilst others will reflect differences in prevalence.

This is nearing completion. 

Objective 5:

To explore relationships between practice consultation rates and hospital admission rates

Objective 6:

To explore relationships between GP workload and QOF performance, patient satisfaction and mortality

The entire dataset of consultations for the year 2103/2014 were analysed to address these objectives.  Relationships between general practitioner (GP) and nurse consultation rates and outcomes were investigated using negative binomial and ordinal logistic regression models.

These objectives have been written up together and are currently under consideration for publication in BMC Health Services Research.

Plain English summary

General practice is central to provision of care in the NHS. Yet we know very little about what goes on in general practice – who consults, how often, or with what sorts of problems. Most of the information we have comes from the Fourth national morbidity study, conducted more than 20 years ago. Since then general practice has changed hugely. The aim of this study was to describe the types of problems that are presented to GPs and practice nurses by different groups of patients.

We also wanted to see whether how busy a practice is affects the quality of care provided to patients, patient satisfaction, and use of hospitals.

We used a largedataset of patient records (from 2.7 million people) from general practice linked to routinely available data about practice characteristics, patient satisfaction, quality of care, hospital admission rates and death rates in those practices for one year from April 2013 – March 2014.  To understand the problems that people visit their GP practice with, we used a smaller random sample of 304,937 people.  All data were anonymous, so that researchers could not identify any patients or practices. 

We used a coding system to describe the types of problems that people go to their GP practice about.  This system split up all the reasons based on 17 different types of body system (such as breathing, digestive system, ears, eyes etc) and we looked at these in 8 age groups from the very youngest to the oldest and for males and females, separately.  We included 1.17million consultations.

We have shown which types of problems are most common in general practice, such as breathing and skin problems.  We have also shown that a lot of work in doctor’s surgeries is about general issues like checking blood pressure and providing lifestyle advice.  We have been able to look to see how different types of problems are seen in general practice by different age groups, and showing how the youngest and the oldest are seen more often.

We have compared the current patterns that we have seen with the data that was collected over 20 years ago to see how the use of family doctors and nurses has changed over time.

Perhaps reassuringly we have also shown that rates of general practitioner (GP) and nurse consultation are not related to higher hospital admission rates or death rates.  Patient satisfaction and quality of care also do not seem to be affected by high workloads.

Dissemination

Published articles

Three papers of pre-curser work to this application have been published (funded by DH).  These were all conducted by the same team and were co-incident with the SPCR funding and have been appropriately badged

1. https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(16)00620-6/fulltext

2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5701995/pdf/bmjopen-2017-018261.pdf

3. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5916084/pdf/bjgpmay-2018-68-670-e370-oa.pdf

Submitted Papers

• Patient consultation rate and clinical and NHS outcomes: a cross-sectional analysis of English primary care data from 2.7 million patients in 238 practices.  S Lay-Flurrie, E Mathieu, C Bankhead, BD Nicholson, R Perera-Salazar, T Holt, FDR Hobbs, C Salisbury, on behalf of the NIHR School for Primary Care Research, Nuffield Department of Primary Care Health Sciences, University of Oxford.  Submitted to BMC Health services Research.

Papers in Preparation

• Morbidity statistics from general practice, a fifth national study: cross-sectional analysis.  C Bankhead, A Fuller, FDR Hobbs, T Holt, S Lay-Flurrie, BD Nicholson, R Perera-Salazar, C Salisbury, on behalf of the NIHR School for Primary Care Research, Nuffield Department of Primary Care Health Sciences, University of Oxford.

We have also been examining the complexity of consultations and have developed a measure of complexity, since workload is related not only to the number of consultations but also the complexity of the problems encountered.  This work is likely to be reported in two papers:

• The development and validation of a complexity score for general practice;
• The changing complexity of general practice. 

Whilst this work has been conducted with salary support from the Oxford Biomedical Research Centre, it has been conducted by the same core team (CB, AF, SL-F, FDRH, TH, BDN,RP-S, CS), with additional expertise from other collaborators,  and has utilised the data from this research.  Therefore, these will also be badged under this research grant.

 

Public involvement

No public or patient involvement has been conducted , as explained in the original application .

Impact

The main paper that formed the pre-curser to this research, conducted by the same team and seamlessly with this research, has been highly cited and informed policy because it provides the strongest and most up-to-date evidence of rising workload in general practice. We anticipate that the subsequent papers will also be influential because they will demonstrate the content of this workload, how it has changed over the last 20 years, and the primary care needs of different groups of the population

This project was funded by the National Institute for Health Research School for Primary Care Research (project number 243)

Department of Health Disclaimer

The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the NIHR School for Primary Care Research, NIHR, NHS or the Department of Health.