Do care plans and annual reviews of physical health influence unplanned hospital utilisation for people with serious mental illness? Analysis of linked longitudinal primary and secondary healthcare records in England

Objective To investigate whether two primary care activities that are framed as indicators of primary care quality (comprehensive care plans and annual reviews of physical health) influence unplanned utilisation of hospital services for people with serious mental illness (SMI). Design, setting, participants Retrospective observational cohort study using linked primary care and hospital records (Hospital Episode Statistics) for 5158 patients diagnosed with SMI between April 2006 and March 2014, who attended 213 primary care practices in England that contribute to the Clinical Practice Research Datalink GOLD database. Outcomes and analysis Cox survival models were used to estimate the associations between two primary care quality indicators (care plans and annual reviews of physical health) and the hazards of three types of unplanned hospital utilisation: presentation to accident and emergency departments (A&E), admission for SMI and admission for ambulatory care sensitive conditions (ACSC). Results Risk of A&E presentation was 13% lower (HR 0.87, 95% CI 0.77 to 0.98) and risk of admission to hospital for ACSC was 23% lower (HR 0.77, 95% CI 0.60 to 0.99) for patients with a care plan documented in the previous year compared with those without a care plan. Risk of A&E presentation was 19% lower for those who had a care plan documented earlier but not updated in the previous year (HR: 0.81, 95% CI 0.67 to 0.97) compared with those without a care plan. Risks of hospital admission for SMI were not associated with care plans, and none of the outcomes were associated with annual reviews. Conclusions Care plans documented in primary care for people with SMI are associated with reduced risk of A&E attendance and reduced risk of unplanned admission to hospital for physical health problems, but not with risk of admission for mental health problems. Annual reviews of physical health are not associated with risk of unplanned hospital utilisation.


Strengths and limitations of this study
• Innovative use of linked individual-level patient data from primary care, inpatient admissions and A&E, allowing the sequence of care indicators and unplanned hospital utilisation to be identified.
• Contributes to a limited evidence base on the association of patient outcomes with the types of care incentivised under a national primary care physician incentive scheme.
• Due to the observational nature of the data, no information was available on the circumstances leading to documentation of care quality indicators, or their quality or content.
• Partial capacity to account for patient health status and factors driving utilisation of primary and secondary care.

Keywords
Schizophrenia & psychotic disorders; primary care; quality in health care; data linkage; hospital utilisation.

Funding statement
This project was funded by the National Institute for Health Research HS&DR programme (project number 13/54/40). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health.

Competing interests
No competing interests. costs to society. It is linked with a high disease burden 1 2 , poor health outcomes, high treatment costs, and lower life expectancy, primarily attributed to preventable physical causes [3][4][5][6] . Patients with SMI have high rates of A&E attendance and hospital admission, for both physical and mental health problems [7][8][9][10] .
In the UK, while specialist mental health services are important to many patients' care, general practice provides the majority of care for patients with SMI 11 . High quality primary care therefore has the potential to improve the management, health and wellbeing of patients with SMI, and can help meet the aims of policymakers seeking to improve population wellbeing and contain costs 12 13 .
Several countries have adopted incentive programs to improve the quality and value of healthcare 14 , despite mixed evidence of their effectiveness [15][16][17] . England  There is little evidence on whether these incentivised activities improve patients' mental or physical health, modify patterns of healthcare utilisation, or reduce costs.
They derive from NICE guideline recommendations for primary care management of SMI, which are based on consensus in the absence of robust evidence 19 20 . As the indicators are regularly reviewed and revised, new evidence on their potential impacts is useful for decision-makers. Our study contributes such evidence, examining whether care plans and annual reviews are associated with utilisation of unplanned secondary care for patients with SMI, by analysing linked primary care and hospital records.

Quality indicators
A good quality care plan documents the patient's current health and social care needs, and how these will be addressed, in agreement with patients and caregivers. It should specify arrangements with secondary care services (where applicable), and recommendations in case of relapse, including the patient's care preferences and goals 19 20 . The QOF incentivises GPs to document care plans annually, so we analyse 'current' care plans, those recorded in the last twelve months. However, we hypothesise that an older care plan may still influence patients' health service utilisation, and therefore also analyse 'expired' care plans, those recorded more than a year ago.
The annual review of physical health indicator changed over the study period. blood pressure, body mass index (BMI), blood glucose or glycated haemoglobin, and ratio of total to high-density lipoprotein cholesterol, and, where appropriate, cervical screening. Unlike care plans, only 'current' annual reviews (recorded within the last twelve months) are included in the analysis, since expired monitoring of physical health is unlikely to have ongoing benefits.
A previous study using data aggregated to practice level found that higher rates of SMI admission were associated with higher achievement on the annual physical review indicator, but not the care plan indicator 21 . However, that study could not ascertain which individuals within a practice received care or whether the care preceded or followed admissions. An innovation of the current study is that we observe the sequence of events for individual patients (dates of SMI diagnosis, care indicators, and hospital utilisation), which allows us to identify whether the care indicator precedes the outcome of interest.
We hypothesise that care plans and annual reviews may modify risk of three outcomes: presentations to Accident and Emergency departments (A&E), unplanned hospital admission for SMI, and unplanned admissions for ambulatory care sensitive conditions (ACSCs), which are potentially preventable by high-quality primary care 22 . Patients with SMI are less likely to report health problems than their peers due to self-neglect, impaired motivation, and social withdrawal 23 24 . Proactive care by GPs may help to identify and manage physical and mental health problems without the need for hospital care.

DATA
We analyse linked primary care, inpatient care, and A&E data for patients with a diagnosis of SMI. The main data source is the Clinical Practice Research Datalink (CPRD) GOLD, which holds linked individual-level anonymized primary care records from participating general practices in the UK. These data are representative of the English population with respect to age and gender, but not region. For example, the north-east of England is under-sampled relative to areas in the west and south 25 .
The CPRD data service provided information on all patients eligible for linkage and registered with a participating practice in England, and with a diagnosis of SMI documented on or before 31 March 2014. Most of the primary care information is recorded using a hierarchical coding system known as 'Read codes' 26 , which we use to identify records of care quality indicators, morbidity profiles, and smoking and alcohol consumption status.

Data linkages
The Hospital Episode Statistics (HES) comprise detailed records for all NHS patients admitted to hospital and presentations to A&E in England. CPRD provided deterministically linked admission and A&E data for all patients included in the study.
Unplanned admissions were classified for SMI by primary diagnosis International Classification of Diseases (ICD-10) codes, categories F20-F31 27 , and for chronic ACSC as defined by Bardsley, et al. 28 . We specified chronic ACSCs since these would be most amenable to the types of GP care we investigate and are a proxy for avoidable admissions 29 . A&E presentations of any cause (mental or physical health) were included. those whose unobserved past care or events could influence their subsequent care. We use data from the year prior to diagnosis to capture elements of medical history.
Following clinical advice, we constructed three diagnostic categories: schizophrenia and other psychoses (group 1), bipolar disorder and affective psychoses (group 2) 1, 2 , and those who had a diagnosis from each group applied at some point. (See   Supplementary Table 1 for the list of codes used to identify each category.) In order to explore annual reviews for the full period of our analysis, given the changes to the annual review incentives in 2011/12, we formulate an aggregate indicator. This signifies that the patient had at least three of four 'health risk' checks (blood pressure, BMI, cholesterol, and glucose) documented within a three month

Study design
We investigate the relationship between having a current or expired care plan or a current annual physical health review, and the occurrence of three separate outcomes: A&E presentations, SMI admissions, and ACSC admissions. We use duration analysis to examine how time to an outcome is affected by receipt of care plans and annual reviews. Our approach accommodates the time-varying nature of care plans and annual reviews, which do not necessarily occur at consistent intervals after diagnosis.

Patient involvement
Two co-authors on the multidisciplinary team responsible for this study had lived experience of SMI. They contributed to the design of the research questions and methodological approach, interpretation of the findings, and writing the paper.

Analytical approach
We estimate Cox proportional hazard models 31 with the same basic structure, examining duration from SMI diagnosis to each of the three outcomes: first A&E presentation, first SMI admission, and first ACSC admission. We follow each individual until the outcome of interest or until censoring. Censoring can occur We control for demographic variables at baseline including gender, ethnicity, and age at first diagnosis (the start of the observation period). To account for patients' state of health at entry to the study, we include the number of Charlson comorbidities 32 listed in their primary care records, and whether a diagnosis of depression was recorded.
Patients' history of alcohol consumption and smoking are classified as binary variables (ever/ never) according to whether a relevant diagnostic or management code was ever recorded in CPRD GOLD before the first SMI diagnosis. Historical healthcare utilisation is captured by the number of GP contacts (face-to-face visits and telephone calls) and the number of inpatient admissions (all-cause) in the year preceding diagnosis, and we control for the calendar year of first diagnosis.

Survival analysis
We estimate the following hazard function for each outcome, which is the product of the baseline hazard, ߣ ሺ‫ݐ‬ሻ, and an exponentiated linear function of patients' characteristics and three care indicator variables for current and expired care plans and (current only) annual physical reviews:

Sensitivity analyses
We undertake five tests of the sensitivity of our findings to alternate specifications of the model. First, we test the robustness of the annual review variable, which in our main analysis is the composite indicator constructed from both the original annual review codes and the specific health checks that contribute to such reviews. We

RESULTS OF DURATION ANALYSIS
Results for each of the outcomes are presented in Figure 1. Since our research question addresses the relationship between care quality indicators and outcomes, we focus on the HRs for the three key explanatory variables: current annual review, current care plan and expired care plan. HRs for other explanatory variables are presented in Supplementary Table 3.

Principal findings
Our study examines the relationship between two primary activities framed as care quality indicators (care plans and annual physical health reviews) for patients with SMI, and three measures of secondary healthcare utilization: A&E presentations, SMI admissions, and ACSC admissions. Although care plans and annual physical health reviews were incentivised by the QOF we do not test the effect of incentives. Our research question is whether care plans or annual reviews are influence unplanned hospital utilisation. We find that care plans are associated with a lower hazard of A&E presentation, but not with risk of hospital admissions for SMI or ACSC, and we find no association between annual reviews and the hazard of any of the outcomes.

Strengths and weaknesses
Our study makes a number of important contributions to the evidence base, being (to our knowledge) the first to use linked patient-level data to investigate the relationship between incentivized primary care quality and hospital care for people with SMI in the English NHS. By linking datasets, we track individual patients across primary care and hospital settings over a number of years, and determine the sequence of GP care and subsequent hospital utilisation. Our duration analysis exploits this, providing more robust estimates of the association between care quality indicators and patient's healthcare utilisation than would be possible with aggregate data.
There are a number of limitations to our study. The measures of health status and healthcare utilisation prior to diagnosis of SMI may not fully depict the complexities However, our findings were robust to whether the number of comorbidities was fixed at baseline or allowed to vary over the period of observation. The outcomes we analyse are measures of hospital utilisation, which may act as imperfect proxies for health outcomes. Our finding of a negative association between care plans and A&E presentations may reflect the presence of unobserved factors that could contribute to the observed association without implying causality. We lack information on the clinical circumstances leading to a care plan or annual review being recorded, and on its content, quality and appropriateness. In the period preceding an A&E attendance, care planning may be superseded by acute management of physical or mental health conditions. Patients who have less insight or lower levels of self-care may be less likely to seek regular primary care, and therefore less likely to have care plans documented, but may also be more likely to seek A&E care. Conversely, our findings could underestimate any real effect of care plans on A&E presentations if other unobserved factors contribute to a (non-causal) positive association, such as care plans being triggered by patients attending their GP for problems that will eventually need hospital care.
We restricted the sample to patients registered with the same practice for the year preceding diagnosis in order to analyse a group of patients at a similar clinical stage, and allow us to include measures of historical utilization and medical history.
However, the restriction may have excluded individuals with more severe SMI if this led them to move. In addition, the sample may include some patients who were not newly diagnosed, if their diagnostic information had not been transferred to a new practice, or to a new electronic recording system.

Interpretation of results and implications for clinicians and policymakers
The association of care plans with reduced hazard of A&E presentations but not with that of hospital admissions might suggest that care plans help patients avoid A&E for conditions that do not require hospital care. The documentation of patients' current health status, early warning signs, triggers, social support needs, co-ordination arrangements with secondary care, and preferred course of action in the event of a clinical relapse could improve the management of the patient's health overall and prevent deterioration, reducing the need for urgent care. It may also direct patients into appropriate services during periods of deterioration, and thereby avert A&E presentations.
Increasing demand for A&E care, especially for problems that could be managed elsewhere, is a policy focus in the UK NHS and in other health care systems 33 .
Patients with SMI, in particular, can require high resource input when attending A&E 34 35 . The potential for care planning by GPs to reduce A&E attendances in this patient population could therefore be an important finding for policymakers, suggesting that resources deployed in primary care may reduce demand on secondary care services.

Unanswered questions and future research
This study did not account for care provided by patients' key workers or a crisis resolution and home treatment team, which may provide an alternative to both hospital admission and primary care. The modelling approach does adjust for differences across practices, which may help control for some of these local effects.
To address this more adequately, the Mental Health Services Dataset, which can now be linked to CPRD primary care data, could be used to cover mental health care delivered in the community.

Conclusion
Provider incentive schemes are increasingly popular levers for improving value and quality in healthcare worldwide, but evidence is needed on the effectiveness of incentivised activities. Despite the vulnerability of people with SMI to poor health outcomes, relatively few studies have examined the impact of care quality indicators in this population. This study addresses that evidence gap, and advances our understanding of how primary care can influence utilisation of hospital care in this patient population. We find that care plans are associated with reduced risk of A&E presentation, but not of hospital admissions, supporting the hypothesis that this type of care incentivized under the QOF is achieving at least some of its policy objectives.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   23   Author contributions   JR, PK, NG, CK, TD, AM, NR, HG, MG, TK, NS, SG, CD, LA, RW and RJ   contributed to the design of the research questions, analytical approach, and interpretation of the findings. JR produced a first draft of the paper and contributed to and PK, NG, CK, TD, AM, NR, HG, MG, TK, NS, SG, CD, LA, RW and RJ approved the final manuscript.

Data sharing
Due to the sensitive and confidential nature of the data used for this analysis, and the permissions required to access it, the dataset is not publicly available.

Strengths and limitations of this study
• Innovative use of linked individual-level patient data from primary care, inpatient admissions and A&E, allowing the sequence of care indicators and unplanned hospital utilisation to be identified.
• Contributes to a limited evidence base on the association of patient outcomes with the types of care incentivised under a national primary care physician incentive scheme.
• Due to the observational nature of the data, no information was available on the circumstances leading to documentation of care quality indicators, or their quality or content.
• Partial capacity to account for patient health status and factors driving utilisation of primary and secondary care.

Keywords
Schizophrenia & psychotic disorders; primary care; quality in health care; data linkage; hospital utilisation.

Funding statement
This project was funded by the National Institute for Health Research HS&DR programme (project number 13/54/40). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health.

Competing interests
No competing interests. In the UK, while specialist mental health services are important to many patients' care, general practice provides the majority of care for patients with SMI 11 . High quality primary care therefore has the potential to improve the management, health and wellbeing of patients with SMI, and can help meet the aims of policymakers seeking to improve population wellbeing and contain costs 12 13 .
Several countries have adopted incentive programs to improve the quality and value of healthcare 14  There is little evidence on whether these incentivised activities improve patients' mental or physical health, modify patterns of healthcare utilisation, or reduce costs.
They derive from NICE guideline recommendations for primary care management of SMI, which are based on consensus in the absence of robust evidence 19 20 .
A previous study using data aggregated to practice level found that higher rates of SMI admission were associated with higher achievement on the annual physical review indicator, but not the care plan indicator 21 . However, that study could not ascertain which individuals within a practice received care or whether the care preceded or followed admissions.
As the QOF indicators are regularly reviewed and revised, new evidence on their potential impacts is useful for decision-makers. Our study contributes such evidence, examining whether care plans and annual reviews are associated with utilisation of unplanned secondary care for patients with SMI, by analysing linked primary care and hospital records. We hypothesise that care plans and annual reviews may modify risk of unplanned hospital utilisation, which is potentially preventable by high-quality primary care 22 . Patients with SMI are less likely to report health problems than their peers due to self-neglect, impaired motivation, and social withdrawal 23 24 . Proactive care by GPs may help to identify and manage physical and mental health problems without the need for hospital care.

Study design
In this retrospective observational cohort study using linked primary care and hospital data, we investigate the relationship between having a care plan and/or an annual physical health review, and time from diagnosis of SMI to unplanned hospital utilisation.

Data sources and linkage
The main data source is the Clinical Practice Research Datalink (CPRD) GOLD, which holds linked individual-level anonymized primary care records from participating general practices in the UK. These data are representative of the English population with respect to age and gender, but not region. For example, the north-east of England is under-sampled relative to areas in the west and south 25 . The CPRD data service provided information on all patients eligible for linkage and registered with a participating practice in England, and with a diagnosis of SMI documented on or before 31 March 2014 recorded in clinical notes or referral records, using the event date entered by the GP. The sample was limited to patients whose records met an acceptable standard based on recording of registration, clinical events, and demographic details, at practices deemed up to standard according to a CPRD algorithm 25 . Most of the primary care information is recorded using a hierarchical coding system known as 'Read codes' 26 , which we use to identify records of SMI diagnosis, care quality indicators, and morbidity profiles. CPRD records were linked to Hospital Episode Statistics (HES), which comprise detailed records for all NHS patients admitted to hospital and presentations to A&E in England. CPRD provided deterministically linked admission and A&E data for all patients included in the study. The restriction to newly diagnosed patients excludes those whose unobserved past care or events could influence their subsequent care.

Outcomes
We analyse three outcomes taken from HES data: time to i) presentations to Accident

Care quality indicators
Primary care quality indicators were taken from CPRD data, and capture care plans and annual reviews of physical health based on Read codes specified under the QOF to identify those two indicators. A good quality care plan documents the patient's current health and social care needs, and how these will be addressed, in agreement with patients and caregivers. It should specify arrangements with secondary care services (where applicable), and recommendations in case of relapse, including the patient's care preferences and goals 19 20 . The QOF incentivises GPs to document care plans annually, so we analyse 'current' care plans, those recorded in the last twelve months. However, we hypothesise that an older care plan may still influence patients' health service utilisation, and therefore also analyse 'expired' care plans, those recorded more than a year ago. Conversely, only 'current' annual reviews (recorded within the last twelve months) are included in the analysis, since expired monitoring of physical health is unlikely to have ongoing benefits. The choice of the twelvemonth window to determine expiration status is based on the QOF guidance that a care plan or annual review should be reviewed annually 3 .
From 2006/7 to 2010/11 the QOF annual review indicator entailed the patient being given appropriate health promotion and prevention advice. In 2011/12, this broad indicator was split into more specific indicators: a record of alcohol consumption, checks of blood pressure, body mass index (BMI), blood glucose or glycated haemoglobin, and ratio of total to high-density lipoprotein cholesterol, and, where appropriate, cervical screening. In order to explore annual reviews for the full period of our analysis, we formulate an aggregate indicator. This signifies that the patient had at least three of four 'health risk' checks (blood pressure, BMI, cholesterol, and The analysis includes three time-varying care quality indicator variables. For annual reviews, the base case is no annual review of physical health within the last twelve months. For care plans the base case is never having had a care plan. Two binary variables indicate whether a care plan or annual review has been recorded within the last twelve months ('current care plan' and 'current annual review'). If a patient receives a further care plan or annual review within the twelve-month window, the 'current' period is extended accordingly. A third variable indicates that the patient received a care plan more than twelve months ago ('expired care plan') but none in the last twelve months (mutually exclusive with 'current care plan').

Control variables
Using primary care data, we control for patients' age, gender, ethnicity, deprivation profile of their area of residence 30

Statistical analysis
We estimate Cox survival models 32 , examining duration from SMI diagnosis to each of the three outcomes separately: first A&E presentation, first SMI admission, and first ACSC admission. We follow each individual until the outcome of interest or until censoring. Censoring can occur because i) a patient dies, ii) registration with the practice ends, or iii) the study period ends (i.e. the patient is still registered on 31 March 2014). In

Sensitivity analyses
We undertake five tests of the sensitivity of our findings to alternate specifications of the model. First, we test the robustness of the annual review variable, which in our main analysis is the composite indicator constructed from both the QOF-specified new comorbidities. We lastly apply two alternate specifications of practice-level characteristics, instead of stratifying on practice. The first assumes no practice-level differences in baseline hazard, but includes practice characteristics (rurality and distance from closest acute hospital and inpatient psychiatric unit) as covariates. The second includes practice fixed effects as explanatory variables. Both models adjust the standard errors for clustering at the practice level as before.

Patient involvement
Two co-authors on the multidisciplinary team responsible for this study have lived experience of SMI. They contributed to the design of the research questions and methodological approach, interpretation of the findings, and writing the paper.

Study population
The  Tables 3-5). There were more people per year diagnosed in later years of the study period, but each was observed for a shorter period, and therefore had a shorter window of opportunity to experience both the care quality indicators and the outcomes. Similarly, those aged over 65 made up 25% of the sample but only contributed 18% of the total time observed because they had a shorter than average period of observation (2.01 years). The median age at diagnosis in the sample was 46, and the most common physical comorbidities were respiratory disease, renal disease, cancer, and diabetes (further details available in Supplementary Table 6).  Figure 1).

Outcomes
Summary statistics for each of the three outcome variables are presented in Table 2.
During the period of observation, 50% had an A&E presentation, 11% had an SMI admission, and 10% had an ACSC admission.

Results of survival analysis
Results showing the association of the three key explanatory variables with hazard (HR) of each of the outcomes are presented in Figure 1. HRs for control variables are presented in Supplementary Table 7.

Principal findings
We find that, among patients with SMI, those with a care plan documented in primary care in the last 12 months have a 14% lower hazard of A&E presentation than those without a care plan, and a 23% lower hazard of admission to hospital for ACSC (the types of condition thought to be amenable to primary care). Those with a care plan documented more than 12 months ago also had a lower hazard of A&E presentation (19% lower than those with no care plan). We find no such association between annual reviews of physical health and the hazard of any of the outcomes, nor between care plans and admissions for SMI.

Strengths and weaknesses
Our study makes a number of important contributions to the evidence base, being (to our knowledge) the first to use linked patient-level data to investigate the relationship between incentivized primary care quality and hospital care for people with SMI in the English NHS. By linking datasets, we track individual patients across primary care and hospital settings over a number of years, and determine whether the primary care indicator precedes the hospital utilisation. Our survival analysis exploits this, providing more robust estimates of the association than would be possible with aggregate data. However, there is no reason to consider that the capture of attendances is related to whether a patient had a care plan or physical review, meaning our comparative analysis should not be adversely impacted.

Comparison with other studies
There is limited, mixed evidence about the relationship between objective measures of primary care quality and A&E attendance. Baker, et al. 35 found that higher performance on the QOF overall was not a predictor of area-level A&E attendances, while subjective assessment of higher primary care quality has been found to be associated with lower utilisation of A&E 36 37 . Using practice-level data, Gutacker, et al. 21 found that higher achievement on the annual review indicator was associated with a higher rate of SMI admissions. One possible explanation was that patients had QOF indicators documented after a hospital admission. Our analysis, which captures the sequence of events and finds no association between annual reviews and hazard of SMI admission, supports that explanation. Harrison, et al. 38 found, using aggregate data, that the introduction of the QOF was associated with a decrease in unplanned admissions for incentivised ACSCs, consistent with our finding that care plans were associated with a reduced risk of ACSC admissions. Wilson, et al. 39 found, using individual-level data, that while introduction of QOF incentives increased detection of cardiovascular risk factors in patients with SMI, there was not necessarily any change in management. This is consistent with our finding no evidence of improved health outcomes (or their proxies, unplanned hospital utilisation) associated with annual reviews of physical health.

Interpretation of results and implications for clinicians and policymakers
The association of care plans with reduced hazard of A&E presentations and ACSC

Unanswered questions and future research
This study did not account for care provided by patients' key workers or a crisis resolution and home treatment team, which may provide an alternative to both hospital admission and primary care. The modelling approach does adjust for differences across practices, which may help control for some of these local effects.
To address this more adequately, the Mental Health Services Dataset, which can now be linked to CPRD primary care data, could be used to cover mental health care delivered in the community.

Conclusion
Provider incentive schemes are increasingly popular levers for improving value and quality in healthcare worldwide, but evidence is needed on the effectiveness of incentivised activities. Despite the vulnerability of people with SMI to poor health outcomes, relatively few studies have examined the impact of care quality indicators in this population. This study addresses that evidence gap, and advances our

Data sharing
Due to the sensitive and confidential nature of the data used for this analysis, and the permissions required to access it, the dataset is not publicly available.

Survival analysis
We estimate the following hazard function for each outcome, which is the product of the

Strengths and limitations of this study
• Innovative use of linked individual-level patient data from primary care, inpatient admissions and A&E, allowing the sequence of care indicators and unplanned hospital utilisation to be identified.
• Contributes to a limited evidence base on the association of patient outcomes with the types of care incentivised under a national primary care physician incentive scheme.
• Due to the observational nature of the data, no information was available on the circumstances leading to documentation of care quality indicators, or their quality or content.
• Partial capacity to account for patient health status and factors driving utilisation of primary and secondary care.

Keywords
Schizophrenia & psychotic disorders; primary care; quality in health care; data linkage; hospital utilisation.

Funding statement
This project was funded by the National Institute for Health Research HS&DR programme (project number 13/54/40). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health.

Competing interests
No competing interests. In the UK, while specialist mental health services are important to many patients' care, general practice provides the majority of care for patients with SMI 11 . High quality primary care therefore has the potential to improve the management, health and wellbeing of patients with SMI, and can help meet the aims of policymakers seeking to improve population wellbeing and contain costs 12 13 .
Several countries have adopted incentive programs to improve the quality and value of healthcare 14  There is little evidence on whether these incentivised activities improve patients' mental or physical health, modify patterns of healthcare utilisation, or reduce costs.
They derive from NICE guideline recommendations for primary care management of SMI, which are based on consensus in the absence of robust evidence 19 20 .
A previous study using data aggregated to practice level found that higher rates of SMI admission were associated with higher achievement on the annual physical review indicator, but not the care plan indicator 21 . However, that study could not ascertain which individuals within a practice received care or whether the care preceded or followed admissions.
As the QOF indicators are regularly reviewed and revised, new evidence on their potential impacts is useful for decision-makers. Our study contributes such evidence, examining whether care plans and annual reviews are associated with utilisation of unplanned secondary care for patients with SMI, by analysing linked primary care and hospital records. We hypothesise that care plans and annual reviews may modify risk of unplanned hospital utilisation, which is potentially preventable by high-quality primary care 22 . Patients with SMI are less likely to report health problems than their peers due to self-neglect, impaired motivation, and social withdrawal 23 24 . Proactive care by GPs may help to identify and manage physical and mental health problems without the need for hospital care.

Study design
In this retrospective observational cohort study using linked primary care and hospital data, we investigate the relationship between having a care plan and/or an annual physical health review, and time from diagnosis of SMI to unplanned hospital utilisation.

Data sources and linkage
The main data source is the Clinical Practice Research Datalink (CPRD) GOLD, which holds linked individual-level anonymized primary care records from participating general practices in the UK. These data are representative of the English population with respect to age and gender, but not region. For example, the north-east of England is under-sampled relative to areas in the west and south 25 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   8 To preserve anonymity, all linkages were carried out under the CPRD routine linkage scheme.

Inclusion and exclusion criteria
Our The restriction to newly diagnosed patients excludes those whose unobserved past care or events could influence their subsequent care.

Care quality indicators
Primary care quality indicators were taken from CPRD data, and capture care plans and annual reviews of physical health based on Read codes specified under the QOF to identify those two indicators. A good quality care plan documents the patient's current health and social care needs, and how these will be addressed, in agreement with patients and caregivers. It should specify arrangements with secondary care services (where applicable), and recommendations in case of relapse, including the patient's care preferences and goals 19 20 . The QOF incentivises GPs to document care plans annually, so we analyse 'current' care plans, those recorded in the last twelve months. However, we hypothesise that an older care plan may still influence patients' health service utilisation, and therefore also analyse 'expired' care plans, those recorded more than a year ago. Conversely, only 'current' annual reviews (recorded within the last twelve months) are included in the analysis, since expired monitoring of physical health is unlikely to have ongoing benefits. The choice of the twelvemonth window to determine expiration status is based on the QOF guidance that a care plan or annual review should be reviewed annually 3 .
From 2006/7 to 2010/11 the QOF annual review indicator entailed the patient being given appropriate health promotion and prevention advice. In 2011/12, this broad indicator was split into more specific indicators: a record of alcohol consumption, checks of blood pressure, body mass index (BMI), blood glucose or glycated haemoglobin, and ratio of total to high-density lipoprotein cholesterol, and, where appropriate, cervical screening. In order to explore annual reviews for the full period of our analysis, we formulate an aggregate indicator. This signifies that the patient had at least three of four 'health risk' checks (blood pressure, BMI, cholesterol, and  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   10 glucose) documented within a three month period 29 . The date of the aggregate indicator is the date of the final check. In our analysis, patients are considered to have a 'current' annual review for twelve months after either an annual review (using the original Read coding), or from the date of this aggregate indicator.
The analysis includes three time-varying care quality indicator variables. For annual reviews, the base case is no annual review of physical health within the last twelve months. For care plans the base case is never having had a care plan. Two binary variables indicate whether a care plan or annual review has been recorded within the last twelve months ('current care plan' and 'current annual review'). If a patient receives a further care plan or annual review within the twelve-month window, the 'current' period is extended accordingly. A third variable indicates that the patient received a care plan more than twelve months ago ('expired care plan') but none in the last twelve months (mutually exclusive with 'current care plan').

Patient involvement
Two co-authors on the multidisciplinary team responsible for this study have lived experience of SMI. They contributed to the design of the research questions and methodological approach, interpretation of the findings, and writing the paper.

Study population
The full sample consists of 5,158 newly diagnosed patients with SMI from 213 practices in England with a total of 14,376 person-years observed before censoring (for A&E analysis 4,446 patients and a total of 10,952 person-years). On average patients were observed for 2.79 years. Most (67%) were still observed at the end of the study period, 23% exited because their registration with the practice ended during the study period, and 10% died. Table 1 presents the number of individuals with each  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   13 characteristic, and the contribution of those with each characteristic to the total time observed. (Equivalent tables are presented for each analysis sample in Supplementary   Tables 3-5). There were more people per year diagnosed in later years of the study period, but each was observed for a shorter period, and therefore had a shorter window of opportunity to experience both the care quality indicators and the outcomes. Similarly, those aged over 65 made up 25% of the sample but only contributed 18% of the total time observed because they had a shorter than average period of observation (2.01 years). The median age at diagnosis in the sample was 46, and the most common physical comorbidities were respiratory disease, renal disease, cancer, and diabetes (further details available in Supplementary Table 6).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Overall, 69% of the sample had a current care plan at least once, contributing 40% of the total time observed, while 72% of the sample had a current annual review at least once, contributing 60% of the total time observed, and 42% of the sample had an expired care plan at least once, contributing 28% of the total time observed. More detail on the annual rates of care plan and annual review indicators is available in supplementary material (Supplementary Figure 1).

Outcomes
Summary statistics for each of the three outcome variables are presented in Table 2.

Results of survival analysis
Results showing the association of the three key explanatory variables with hazard (HR) of each of the outcomes are presented in Figure 1. HRs for control variables are presented in Table 3 with A&E as the outcome.

Principal findings
We find that, among patients with SMI, those with a care plan documented in primary care in the last 12 months have a 13% lower hazard of A&E presentation than those without a care plan, and a 23% lower hazard of admission to hospital for ACSC (the types of condition thought to be amenable to primary care). Those with a care plan documented more than 12 months ago also had a lower hazard of A&E presentation (19% lower than those with no care plan). We find no such association between annual reviews of physical health and the hazard of any of the outcomes, nor between care plans and admissions for SMI.

Strengths and weaknesses
Our study makes a number of important contributions to the evidence base, being (to our knowledge) the first to use linked patient-level data to investigate the relationship between incentivized primary care quality and hospital care for people with SMI in the English NHS. By linking datasets, we track individual patients across primary care and hospital settings over a number of years, and determine whether the primary care indicator precedes the hospital utilisation. Our survival analysis exploits this, providing more robust estimates of the association than would be possible with aggregate data. There are a number of limitations to our study. The measures of health status and healthcare utilisation prior to diagnosis of SMI may not fully depict the complexities of health status, including severity of SMI, and this may confound the results.
However, our findings were robust to whether the number of comorbidities was fixed at baseline or allowed to vary over the period of observation. The outcomes we analyse are measures of hospital utilisation, which are imperfect proxies for health outcomes. Our finding of a negative association between care plans and A&E presentations or ACSC admissions may reflect the presence of unobserved factors that could contribute to the observed association without implying causality. We lack information on the clinical circumstances leading to a care plan or annual review being recorded, and on its content, quality and appropriateness. In the period preceding hospital attendance, it is possible that care planning could be superseded by acute management of physical or mental health conditions. Patients who have less insight or lower levels of self-care may be less likely to seek regular primary care, and therefore less likely to have care plans documented, but may also be more likely to seek A&E care 33 34 . Conversely, our findings could underestimate any real effect of care plans on A&E presentations if other unobserved factors contribute to a (noncausal) positive association, such as care plans being triggered by patients attending their GP for problems that will eventually need hospital care.
We restricted the sample to patients registered with the same practice for the year preceding diagnosis in order to analyse a group of patients at a similar clinical stage, and allow us to include measures of historical utilization and medical history.
However, the restriction may have excluded individuals with more severe SMI if this However, there is no reason to consider that the capture of attendances is related to whether a patient had a care plan or physical review, meaning our comparative analysis should not be adversely impacted.

Comparison with other studies
There is limited, mixed evidence about the relationship between objective measures of primary care quality and A&E attendance. Baker, et al. 35 found that higher performance on the QOF overall was not a predictor of area-level A&E attendances, while subjective assessment of higher primary care quality has been found to be associated with lower utilisation of A&E 36 37 . Using practice-level data, Gutacker, et al. 21 found that higher achievement on the annual review indicator was associated with a higher rate of SMI admissions. One possible explanation was that patients had QOF indicators documented after a hospital admission. Our analysis, which captures the sequence of events and finds no association between annual reviews and hazard of SMI admission, supports that explanation. Harrison, et al. 38 found, using aggregate data, that the introduction of the QOF was associated with a decrease in unplanned admissions for incentivised ACSCs, consistent with our finding that care plans were associated with a reduced risk of ACSC admissions. Wilson, et al. 39 found, using individual-level data, that while introduction of QOF incentives increased detection of cardiovascular risk factors in patients with SMI, there was not necessarily any change in management. This is consistent with our finding no evidence of improved health outcomes (or their proxies, unplanned hospital utilisation) associated with annual reviews of physical health.

Interpretation of results and implications for clinicians and policymakers
The association of care plans with reduced hazard of A&E presentations and ACSC admissions might suggest that care plans help patients avoid hospital for conditions that do not require hospital care. The documentation of patients' current health status, early warning signs, triggers, social support needs, co-ordination arrangements with secondary care, and preferred course of action in the event of a clinical relapse could improve the management of the patient's health overall and prevent deterioration, reducing the need for urgent care represented by ACSC admissions and A&E presentations. It may also direct patients into appropriate services during periods of deterioration, and thereby avert hospital use.
Increasing demand for A&E care, especially for problems that could be managed elsewhere, is a policy focus in the UK NHS and in other health care systems 40 .
Patients with SMI, in particular, can require high resource input when attending A&E 41 42 . The potential for care planning by GPs to reduce A&E attendances in this patient population could therefore be an important finding for policymakers, suggesting that continued incentivisation of this activity in primary care may help reduce demand on secondary care services. The associations of both 'current' and 'expired' care plans with lower hazard of A&E attendance suggest that an older care plan can still be beneficial. However, the finding that when care plans are considered effective for 15 months rather than 12 months, only 'current' care plans are associated with reduced A&E attendance suggests that there are limits to the duration of impact. The lack of association between annual reviews of physical health and unplanned hospital care may suggest that effectiveness of a GP in managing a patient's health, and preventing the need for A&E or hospital care, may be unaffected by whether or not annual reviews or health checks are recorded.

Unanswered questions and future research
This study did not account for care provided by patients' key workers or a crisis resolution and home treatment team, which may provide an alternative to both hospital admission and primary care. The modelling approach does adjust for differences across practices, which may help control for some of these local effects.
To address this more adequately, the Mental Health Services Dataset, which can now be linked to CPRD primary care data, could be used to cover mental health care delivered in the community.

Conclusion
Provider incentive schemes are increasingly popular levers for improving value and quality in healthcare worldwide, but evidence is needed on the effectiveness of incentivised activities. Despite the vulnerability of people with SMI to poor health outcomes, relatively few studies have examined the impact of care quality indicators in this population. This study addresses that evidence gap, and advances our

Data sharing
Due to the sensitive and confidential nature of the data used for this analysis, and the permissions required to access it, the dataset is not publicly available.

Ethical approval
The study protocol was approved by the Independent Scientific Advisory Committee of the Clinical Practice Research Datalink (protocol number 15_213R2A2).