Article Text
PDF
Abstract
Background Physicians’ work conditions and mental well-being may affect healthcare quality and efficacy. Yet the effects on objective measures of healthcare performance remain understudied. This study examined mental well-being, job satisfaction and self-rated workability in general practitioners (GPs) in relation to hospitalisations for ambulatory care sensitive conditions (ACSC-Hs), a register-based quality indicator affected by referral threshold and prevention efforts in primary care.
Methods This is an observational study combining data from national registers and a nationwide questionnaire survey among Danish GPs. To ensure precise linkage of each patient with a specific GP, partnership practices were not included. Study cases were 461 376 adult patients listed with 392 GPs. Associations between hospitalisations in the 6-month study period and selected well-being indicators were estimated at the individual patient level and adjusted for GP gender and seniority, list size, and patient factors (comorbidity, sociodemographic characteristics).
Results The median number of ACSC-Hs per 1000 listed patients was 10.2 (interquartile interval: 7.0–13.7). All well-being indicators were inversely associated with ACSC-Hs, except for perceived stress (not associated). The adjusted incidence rate ratio was 1.26 (95% CI 1.13 to 1.42) for patients listed with GPs in the least favourable category of self-rated workability, and 1.19 (95% CI 1.05 to 1.35), 1.15 (95% CI 1.04 to 1.27) and 1.14 (95% CI 1.03 to 1.27) for patients listed with GPs in the least favourable categories of burn-out, job satisfaction and general well-being (the most favourable categories used as reference). Hospitalisations for conditions not classified as ambulatory care sensitive were not equally associated.
Conclusions ACSC-H frequency increased with decreasing levels of GP mental well-being, job satisfaction and self-rated workability. These findings imply that GPs’ work conditions and mental well-being may have important implications for individual patients and for healthcare expenditures.
- general practice
- human factors
- quality improvement
- ambulatory care
- primary care
Statistics from Altmetric.com
Introduction
Mental distress, such as stress and burn-out, is increasingly common in physicians, including general practitioners (GPs).1–3 Poor mental well-being and low job satisfaction may have significant negative implications for the provision of healthcare.4–8 Compared with physicians with good mental well-being and high job satisfaction, physicians with poor mental well-being and little job satisfaction report lower levels of job performance.5 6 8 This could reflect a negative self-image influenced by the mental health status rather than actual differences in performance.5 8–10 Few empirical studies have explored physician mental well-being and satisfaction in relation to objective rather than self-reported measures of healthcare performance.5 6
In the Danish healthcare system, GPs play a pivotal role.11 Nearly all residents are listed with a specific general practice, which they must consult for medical advice. GPs provide comprehensive family medicine, including preventive care for chronic diseases and handling of acute problems (which they must deal with on the same day). The GPs also act as gatekeepers to the rest of the healthcare system (except for life-threatening emergencies). Thus, if mental well-being in GPs affects the provision of primary healthcare, this might have far-reaching implications for both the patients and the cost-efficiency in healthcare.
One possible manifestation of GP distress could be higher use of hospital admissions for ambulatory care sensitive conditions (ACSCs). ACSCs are acute and chronic conditions for which the risk of hospitalisation is reduced by timely and appropriate intervention in primary care.12 The frequency of hospitalisations for ACSC (ACSC-Hs) is considered a valid marker of the quality of primary care.13–15 ACSC-Hs account for considerable healthcare expenses, and prevention of such hospital admissions is a priority in many countries.16
The preventive potential in primary care depends on patients’ access to primary healthcare and the quality of the healthcare provided. Generally, Danish patients have good access to their GP. Yet, if the GP’s efforts to prevent or treat ACSCs are ineffective, the patient’s health status may deteriorate, which may result in a need for hospital care. In addition, the ACSC-H frequency increases with lower referral threshold in the GP. The efforts required to manage ACSCs without hospital referral may sometimes be complex, time-consuming or emotionally demanding. GPs with poor mental well-being may provide suboptimal preventive care and may have a lower referral threshold due to poor mental health, limited internal resources and little confidence in their own clinical judgement. This study aimed to examine whether patients’ exposure to a GP with poor mental well-being, low job satisfaction and reduced self-rated workability increased patients’ likelihood of ACSC-H (used as a healthcare quality indicator) while taking into account selected patient-related risk factors.
Methods
Study population
The study combined register-based data on listed patients (study cases) with data collected from a questionnaire survey on work conditions, well-being and job satisfaction among GPs in Denmark. Details about the questionnaire survey have been published elsewhere.1
All GPs in Denmark were invited to participate in the survey in May 2016 (response rate: 50.2%). For this study, only GPs organised in single-handed practices were included. Approximately 27% of GPs in Denmark work in single-handed practices. GPs in partnership practices (ie, two or more GPs sharing the patient list) were not included as a specific patient could not be linked to an individual GP. GPs were excluded if they had locum(s) employed for >20 hours per week (n=18), if they had <500 listed patients (n=10) or if they had been employed in their current practice for less than 1 year (n=8). Information on type of practice, employment period in current practice and employment of locums was retrieved from the questionnaire survey. Information on listed patients was retrieved from the Patient List Register, which contains information about any citizen’s general practice at any given time.
The study cases were patients aged 25 years or older who had been listed at an eligible practice for at least 1 year at the beginning of the study period. Patients listed for less than 1 year (n=43 724) were not included as their ACSC-H risk may depend on the primary healthcare received while listed at another general practice. The 6-month study period started on 1 May 2016. Patients were censored from the study when no longer listed with the GP (due to death, immigration or change of GP). In total, 461 376 patients listed with 392 GPs were included in the analyses.
Variables
The dependent variable was the number of inpatient hospitalisations for ACSCs. In line with previous Danish studies,17 18 we included 12 ACSCs defined by the Agency for Healthcare Research and Quality in the USA.12 To minimise bias due to hospital-related (rather than GP-related) risk of readmission, rehospitalisation for the same ACSC within 30 days of discharge was not counted as a new ACSC-H. For comparison with ACSC-Hs, we also counted inpatient hospitalisations for conditions that were not classified as ACSCs and thus expected to be less dependent on GP performance. All other causes of hospitalisation were included, except for readmission (for any cause) within 30 days of discharge.
Information on hospitalisations in the study period was retrieved from the Danish National Patient Register, which holds administrative data on all admissions to Danish hospitals with diagnoses classified according to the International Classification of Diseases, 10th revision (ICD-10).19 The principal discharge diagnosis was used to classify a hospitalisation as an ACSC-H. A detailed description of ICD-10 codes used to define the ACSC-Hs (including exclusion criteria) is available elsewhere.18
Our independent variables of interest were GP mental well-being (burn-out, general well-being, general stress and job-related stress), job satisfaction and self-rated workability. These variables are collectively referred to as well-being indicators in this article. Except for job-related stress, all well-being indicators were assessed by validated rating scales. Mental well-being and job satisfaction were measured by full scales, which have previously shown adequate internal consistency among Danish GPs.1 Workability was measured by a single item, which has been shown to have high consistency with the full scale.20
Burn-out was measured by the Maslach Burnout Inventory - Human Services Survey (MBI-HSS).21 Burn-out is considered to develop because of a long-term imbalance between work-related demands and available resources. The MBI-HSS measures three key dimensions: emotional exhaustion (EE), depersonalisation (DP) and feelings of personal accomplishment (PA). Subscale sum scores reflect the degree of burn-out on each dimension. Subscale scores are often categorised according to normative population scores for healthcare professionals. This allows for comparison of burn-out symptoms (high EE, high DP and low PA) over time and across populations, but the clinically significant levels of burn-out are not defined. In line with a previous study,22 we divided each subscale score into four groups based on quartiles to explore dose–response patterns. For all dimensions, the degree of burn-out increased for each group (the PA subscale was reversed), and scores in the fourth group were located within the category of high burn-out according to the normative cut-off scores. To allow for evaluation of burn-out as a multidimensional construct,23 we calculated a composite burn-out score for which all subscales were given equal status. We summed the score for each subscale group: 1 point assigned for subscale scores in the first group, 2 points for scores in the second group, 3 points for scores in the third group and 4 points for scores in the fourth group. Subsequently, this composite score was categorised into five groups: 3–4 points (corresponding to low levels of burn-out on all subscales), 5–6 points, 7–8 points, 9–10 points and 11–12 points (corresponding to high levels of burn-out on all subscales).
Job satisfaction was measured by the Warr-Cook-Wall Job Satisfaction Scale.24 Respondents rated nine job satisfaction aspects and overall job satisfaction. Job satisfaction was categorised according to quartiles of the sum score.
Job-related stress was assessed by one item. The GPs stated how often they experienced their job as unpleasantly stressful. The response was recoded into three categories: ‘Always/often’, ‘Sometimes’ and ‘Rarely/never’.
Perceived stress in life in general was measured by the 10-item version of Cohen’s Perceived Stress Scale25 and categorised according to quartiles of the sum score. The scale assesses perceived stress based on the frequency with which respondents experience stress-related feelings and thoughts.
General well-being was assessed by the WHO Well-Being Index (WHO-5), which consists of five positively phrased items related to quality of daily life. Well-being was categorised as ‘poor’ for a computed scale score of ≤50 (ie, the recommended cut-off score when using the WHO-5 for screening for depression26), ‘good’ for a score of >70 (ie, above the mean score of the general population) and ‘moderate’ for a score in between.1
Self-rated workability was measured by a single item of the Work Ability Index. Respondents rate their current workability against their lifetime best on an 11-point Likert scale from 0 (‘unable to work’) to 10 (‘workability at its best’).20 Self-rated workability was categorised according to quartiles.
Covariates were chosen a priori for adjustment. Sociodemographic factors and comorbidity are important ACSC-H risk factors at patient level.27 We included gender, age (25–34, 35–44, 45–54, 55–64, 65–74, 75–84 or ≥85 years), ethnic origin (Dane or immigrant/descendent), marital status (married/cohabiting or living alone), level of education according to UNESCO’s International Standard Classification of Education (low: ≤10 years; middle: 11–15 years; or high: >15 years), occupation (in the labour force: employed and students; outside the labour force: unemployed, early retirement pensioner, personal or sick leave; or retired), OECD-modified household income (categorised based on quintiles) and comorbidity assessed by Charlson’s Comorbidity Index (CCI) score (0, 1, 2 or ≥3).28 The CCI score was based on diagnoses registered in the Danish National Patient Register (both inpatient and outpatient contacts) in 2006–2015.19 Included covariates at the GP level were gender, seniority (years as a GP: ≤5, 6–15, 16–25 or ≥25) and list size (using groups defined by quintiles).29 Finally, region was included as a covariate because the overall responsibility for general practice and hospitals in Denmark is anchored in five administrative health regions, and inter-regional variation exists in ACSC-H rates.30
Information on GP gender and seniority was obtained from the questionnaire survey. Information on sociodemographic patient factors (for 2015) was obtained from Statistics Denmark.31 Missing information on level of education (5%) was categorised as unknown. Patients with missing information on any of the other listed covariates were excluded (n=659, 0.1%). The civil registration number, a personal identification number assigned to all Danish citizens, was used to link data at the individual level.32 Data on GP factors were linked to each patient through the GP’s provider number. All personal identifiers were encrypted at Statistics Denmark prior to analysis.1
Analyses
For all patients and for subgroups, we calculated the share of patients with at least one ACSC-H and the corresponding 95% CI. For GPs, we calculated the number of ACSC-Hs per 1000 listed patients with 95% CI.
Associations between the GP well-being indicators and the hospitalisation rates (for ACSCs and for other conditions) were calculated at the individual patient level by use of negative binomial regression with follow-up time as exposure. For each type of hospitalisation, eight unadjusted and eight adjusted analyses were carried out (one for each of the independent variables) using the highest level of GP well-being, job satisfaction and self-rated workability as reference. The adjusted analyses included the covariates described above (GP/practice factors, patient factors and region). Robust variance estimation was used to account for clusters of patients at practice level in both unadjusted and adjusted analyses. Associations were estimated as incidence rate ratios (IRR) with 95% CI. In a sensitivity analysis, separate estimates were calculated by type of ACSC-H (chronic or acute). For each well-being indicator, the number of excess ACSC-Hs associated with suboptimal GP well-being was calculated and presented as number per 100 000 patients listed for 6 months. This number was calculated for each level of suboptimal well-being using the adjusted IRR relative to the actual number of ACSC-Hs for the well-being category of interest divided by the total number of patient days at risk within that category multiplied by 100 000 ((ACSC-Hs × (IRR – 1)/IRR)/risk time × 100 000). A p value of ≤0.05 was considered statistically significant. Analyses were performed using Stata V.15.
Results
During the study period of 6 months, 4835 ACSC-Hs were registered in our study cohort of 461 376 patients (mean follow-up time: 181 days). Patient characteristics and share of patients with at least one ACSC-H are displayed in table 1.
Patient characteristics according to ACSC-Hs
The ACSC for which patients were most often hospitalised was pneumonia (1508, 31.2%), followed by urinary tract infection (1042, 21.6%) (table 2).
Hospitalisations for ACSCs in the study
GP characteristics and well-being are presented in table 3.
Description of the GPs included in the study (n=392)
The results of the regression analyses and the number of excess ACSC-Hs related to suboptimal GP well-being are shown in table 4. Overall, the ACSC-H rate tended to increase with decreasing GP well-being. All well-being indicators, except for perceived stress, were statistically significantly associated with ACSC-Hs in both unadjusted and adjusted analyses. Using GPs in the most favourable category of the well-being indicators as reference, we found that the adjusted IRR was particularly high among patients listed with a GP in the least favourable category of self-rated workability (adjusted IRR=1.26 (95% CI 1.13 to 1.42)). This corresponded to 252 more ACSC-Hs per 100 000 patients listed for 6 months than expected if these patients had the same rate as patients listed with a GP reporting optimal workability. The adjusted IRR was 1.19 (95% CI 1.05 to 1.35) for patients listed with a GP in the least favourable composite burn-out score category and 1.15 (95% CI 1.02 to 1.30) for patients listed with a GP in the second least favourable category. This corresponded to 336 excess ACSC-Hs per 100 000 patients listed for 6 months. Overall, the association between well-being indicators and hospitalisation rate was considerably stronger for ACSCs than for conditions not classified as ACSCs.
Hospitalisations for ACSCs and hospitalisations for other conditions in the practice population in relation to the GP’s well-being, job satisfaction and self-rated workability (each well-being indicator examined separately)
The overall trend was the same for all well-being indicators in the comparison between acute and chronic ACSC-Hs (online supplementary appendix table). However, the association with low personal accomplishment appeared stronger for chronic ACSC-Hs than for acute ACSC-Hs. In contrast, the associations with emotional exhaustion, unpleasant job stress, perceived stress and general well-being appeared strongest for acute ACSC-Hs. Acute ACSC-Hs accounted for most of the total ACSC-Hs, as shown in table 2.
Supplemental material
Discussion
Main findings
Patients listed with a GP showing signs of mental distress had a higher ACSC-H frequency than had patients listed with a GP without mental distress; this trend was seen even after taking into account sociodemographic risk factors. Notably, this relationship had a dose–response pattern across several indicators of GP well-being. The associations were particularly strong for self-rated workability and burn-out, but the ordering of the different indicators should be interpreted with caution as CIs were overlapping.
Hospitalisations for other conditions were not equally associated. This finding suggests that primarily hospitalisations for conditions with prevention potential in primary care are influenced by GP well-being.
Overall, these results indicate that a significant and clinically relevant association exists between the GP’s mental well-being/job satisfaction and the risk of ACSC-H in listed patients.
Strengths and limitations
A key strength of this study was the use of data from national registers combined with survey data. This combination enables precise linkage of each patient to an individual GP, use of a care provision measure based on administrative data (rather than relying on subjectively GP-reported patient outcomes) and adjustment for several potential confounders. Additionally, the national registers provide highly valid data.32 Except for job-related stress, all GP well-being indicators were assessed by validated instruments. The well-being indicators examined are inter-related, but they measure different relevant aspects of well-being and satisfaction. All measures were categorised according to predetermined procedures. The large sample size allowed us to rank GP well-being indicators using multiple categories (rather than dichotomisation), which reduced information loss.
The restricted study period of 6 months minimised potential bias related to changes in GP well-being during the observation period. Since the risk of an ACSC-H is likely to depend on preventive steps taken over a longer period, the exclusion of patients listed with their GP for less than 1 year is an additional strength of the study. Yet more assessment points would have been beneficial, as the mental state of the GP might have altered over time. For instance, the level of perceived stress may fluctuate, whereas burn-out is generally considered a form of chronic distress.23 Therefore, it is likely that GPs with advanced levels of burn-out have experienced symptoms for a longer period.
The study design does not permit causal inference. The assessment of GP well-being indicators prior to the outcome measure reduced the likelihood that reverse causation accounted for the observed correlations. Still, poor well-being and low job satisfaction in GPs might be a consequence of high ACSC-H frequency in patients.
As seen from table 1, some patient characteristics, for example, older age and low socioeconomic position, were strongly associated with an increased risk of ACSC-H. The adjustment for key markers of disease burden and overall health at the individual patient level is a major strength. Yet confounding related to patients’ predisposition for ACSC-H may still occur. For example, the CCI does not include information on ‘never hospitalised’ patients and may not sufficiently adjust for effects of morbidity. To refine the assessment of morbidity, some studies have supplemented with prescription data as a proxy for disease.14 Yet prescription data may also reflect the medical practices of the GPs related to the quality of provided care. Adjusting for medical practices, which may be part of the causal pathway between GP distress and ACSC-H rate, could thus mask associations.
The most frequent ACSC-Hs in this study do not presuppose specific underlying diagnoses, although some ACSC-Hs do (eg, diabetes-related ACSC-Hs). The potential impact of actions taken in primary care on the prevalence and severity of such illnesses holds a risk of overadjustment bias. However, the essential determinants of disease are beyond the control of the GPs, and unequal distribution of specific diseases may confound the results if inadequately accounted for.
Unmeasured characteristics of the GPs and their work conditions hold a risk of confounding the results. Underlying work conditions (rather than GPs’ psychological response to these) could have mediated the observed relationship between poor well-being in GPs and ACSC-H frequency. We adjusted for list size, but other unmeasured factors related to workload, for example, the number of listed patients with high healthcare needs, could have affected GPs’ capability of preventing ACSC-Hs and their occupational well-being.33 Unmeasured local differences in hospital services and community care provision (eg, home-care services) may have confounded the results in opposite directions. If poor community care provision was related to GP distress and increased risk of ACSC-H, this could have resulted in an overestimation of the observed relationship between GP distress and ACSC-H. In contrast, short supply of hospital beds and constrained opportunities for hospital admissions may have influenced GP distress, which could have resulted in an underestimation of our results.
Considering the dose–response pattern across several well-being indicators and the adjustment for relevant covariates, we find that unmeasured confounders are unlikely to fully account for the observed associations. Still all estimated effect sizes should be interpreted cautiously in consideration of the complexity of the relationships observed.
We do not know whether a patient was admitted by his or her GP or through out-of-hours services. However, in both situations, hospitalisation may depend on both the accessibility of the patient’s GP and the quality of care provided by the GP.
To validate the hypothesis that the ACSC-H rate may be influenced by GP well-being and job satisfaction, we assessed whether associations could also be found for hospitalisations for conditions not classified as ACSCs. The weaker associations found for these hospitalisations (considered to be less dependent on the quality of primary care) clearly supported the hypothesis. It should be noted that some lists of ACSCs include a wider range of diagnoses than the list used in this study.16 Hence, some hospitalisations for conditions that were not defined as ACSCs in our study may also depend on the quality of primary care. This could account for the weak association observed between some well-being indicators and the hospitalisation rate for conditions not classified as ACSCs. In addition, the finding that ACSC-H frequency was strongly associated with GPs’ self-rated workability, whereas the frequency of hospitalisations for other conditions was not, supports the use of ACSC-H as a quality indicator.
The generalisability of findings may be constrained since the study population was restricted to patients listed with GPs who were organised in single-handed practices and who responded to the survey. However, the prevalence of burn-out and low job satisfaction among GPs in single-handed practices did not differ from that among GPs in partnership practices.1 Non-responding GPs tended to have higher numbers of comorbid, ageing or deprived patients, and we would thus expect more ASCH-Hs to occur among their patients than among the patients included in this study. Nevertheless, the associations examined may not depend on the GP’s approach to participation or type of practice. Hence, we believe that the findings in this study could also apply to partnership practices and to other countries with similar healthcare systems.
Comparison with existing literature
To our knowledge, this is the first study to examine GP mental well-being in relation to frequency of hospitalisations with a prevention potential in primary healthcare. Our findings add to the collective body of research suggesting that poor well-being and dissatisfaction in healthcare providers may negatively influence the quality of healthcare, whereas high levels of well-being and satisfaction may have a positive impact.4–8
Our finding that occupational well-being may influence the use of secondary care aligns with the results from another GP study, which reported higher referral rates for diagnostic tests and specialist care among GPs with higher levels of burn-out and workload than among GPs under less pressure.34 The authors acknowledged that referral rates per se are not indicators of poor healthcare quality, but they argued that GPs under pressure might make more referrals as an easier way to comply with high pressure in the consultations. This is in agreement with theories on the consequences of burn-out, which suggest an intentional or unintentional hesitation to invest resources in the job as a defence against further depletion.35 GPs who are burned out and under pressure may have a tendency to choose the alternative requiring the least efforts when a clinical task can be approached in different ways, especially when the task is perceived as demanding. In some situations, this may imply that the patient is admitted to hospital.
Some studies support that the increased ACSC-H frequency in patients cared for by GPs with poor mental well-being and low job satisfaction may be linked with lower quality of healthcare provided to prevent a need for hospital care. In a longitudinal study, patients of satisfied physicians adhered better to the recommended care than patients of physicians with low job satisfaction.36 A study found that patients cared for by less burned-out primary care physicians had better blood pressure control.37 Job satisfaction and commitment have also been shown to be associated with better prevention and disease management38 and with better patient-perceived care quality.38 39 In contrast, other studies found no association between physician burn-out and interpersonal quality of patient care as rated by the patients or external observers.40–42 Studies examining patient care assessed by audit data for selected chronic conditions found no associations with physician mental well-being or job satisfaction,10 43 but one of these studies found lower quality of preventive care among physicians reporting high levels of time pressure.10 This implies that adverse work conditions could be a third underlying factor that may affect GPs’ practice style, independently of the individual level of mental distress. In addition, the ability to provide high quality of care appears to be an important driver of physician satisfaction.44–46 Both poor well-being and low job satisfaction in GPs seem to represent ‘red flags’ that indicate suboptimal quality of healthcare, but it remains unclear if the mental state is an essential determinant of quality or merely a marker of underlying adverse work conditions. The mechanisms by which GP work conditions and mental well-being may influence care provision are complex and poorly examined.6
Implications and conclusion
Despite the inherent limitations of the observational design, the results suggest that GPs’ work conditions and well-being do affect the ACSC-H rate and thereby the quality and efficiency of healthcare. Although the associations observed are modest, the potential implications for healthcare finances at the population level are considerable and may intensify further with the demographic shift towards more elderly citizens, especially if the number of distressed GPs continues to rise. The potential implications for the patients affected are disturbing if a higher ACSC-H rate is a marker of undue disease progression because of suboptimal healthcare.
The finding of a dose–response pattern across several well-being indicators suggests that initiatives to reduce distress and to foster high levels of well-being and engagement in GPs may prove efficient in improving the quality and efficiency in healthcare.
In conclusion, mental well-being, satisfaction and self-rated workability in GPs were inversely associated with the ACSC-H rate among listed patients. These findings add considerable weight to the existing knowledge that points to well-being in the healthcare provider as a significant healthcare quality issue.
References
Footnotes
Contributors All authors contributed substantially to the design of the study. KBN performed the statistical analyses in consultation with AHC. KBN wrote the first draft of the article. AFP, FB, AHC and PV assisted in writing and revising the manuscript. All authors read and approved the final manuscript.
Funding This study is funded by the Danish National Research Foundation for Primary Care and by the Danish Health Foundation.
Competing interests None declared.
Patient consent for publication Not required.
Ethics approval The project was approved by the Danish Data Protection Agency (J.no. 2016–41-4648). According to Danish law, approval by the Danish National Committee on Health Research Ethics was not required for this study as no biomedical intervention was performed. Respondents gave their consent to participate by responding to the questionnaire. Personally identifiable information on GPs and patients were recoded and anonymised at Statistics Denmark prior to data analysis.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement No data are available.