Continuity in general practice and hospitalization patterns: an observational study

This observational study was based on data, which are quarterly collected by a large health insurer for reimbursement of healthcare providers. The insurer (AOK Baden-Württemberg, Germany) covers about 45% of the population of about 11 million inhabitants in Baden-Württemberg. The data in this study refer to the year 2017 (1 January until 31 December).

As in previous studies of strong general practice care [9, 11, 12], we included patients if they were adults (aged 18 years or older), lived in the region, had one or more visits to a physician in the region, were insured with AOK (without interruptions in the observed year), and were not registered with other healthcare contracts. Patients in the programme had a known connection to a specific general practitioner, because they provided informed consent before participation in the study. We compared them with all other eligible patients in the observed year, who are not registered in the programme. In the analysis, we linked these non-enrolled patients to the physician, whom they had visited in half or more of their ambulatory care contacts. If no such linkage was possible, they were excluded from analysis. We assumed that the most contacted physician was the general practitioner. A total of 212,203 out of 2,818,295 potential control patients (7.5%) were excluded.

The programme for enhancing strong general practice care is specified in Paragraph 73b of the German Social Code Book Five. The general practitioner (GP) receives 40% additional reimbursement for included patients. The included patients did not have direct financial benefit. Physicians and patients can freely choose whether they enroll in the programme. The following requirements were specified:

All measures were based on health insurance claims data. We operationalized continuity of care in three coefficients, following previous research [16]: the Usual Provider of Care (UPC), the Herfindahl Index (HI), and the Sequential Continuity Index (SCN) (Table 1). We did not use the Bice-Boxerman Continuity of Care Index [16], because it is similar to the Herfindahl Index but would imply a restriction of the data-set. For all three coefficients applies that higher values indicate higher continuity of care. UPC indicates the concentration of contacts with a specific healthcare provider within an episode of care. HI indicates the concentration of contacts within providers across all providers involved in care for a patient. SCN indicates sequential contacts with the same provider for all providers aggregated.

Hospitalizations were operationalized as in previous research of GP-centred care [3] in terms of total hospitalizations (number of hospitalizations per patient), rehospitalizations within 4 weeks (independent of admission diagnoses), and potentially avoidable hospitalizations. The latter was determined on the basis of “ambulatory care-sensitive conditions”, which – based on ICD-10 codes – include the most prevalent conditions for hospitalizations that could have potentially been handled in ambulatory care. The underlying calculus is described in detail elsewhere [17] and summarized in Supplement 1. The three measures are based on count data per patient. The corresponding events may occur not (0), once (1) or more than once (n> 1) in a patient in the observed period.

We also extracted patient age (in years), sex (male, female), nationality (German, other), health insurance (member, family, retired), comorbidity (Charlson Index [18]), living in a nursing home, number of contacts in general practice care, and (for the intervention group) length of participation in the programme (in quarter years) from the data-base.

We calculated descriptive characteristics of patients (as listed above) for patients in the strong general practice care programme (intervention cohort) and patients not in the programme (control cohort). To address the research aims, we used linear mixed models [19] with GEEs (Generalized Estimating Equations) in combination with the Poisson distribution. In all regression models, the patient characteristics listed above, except for number of consultations and time in the programme, were included for correction of differences in composition of the two cohorts. Clustering of patients in GPs and GPs in practices was included in the models. For technical reasons, we had to exclude the patients who dropped out of the programme, because we could not assign those patients to one of the both groups. This drop-out rate was 1,4% in this cohort. The intervention and control patients had a one-year death rate of 0.0173 and 0.0178, respectively. Therefore, we did not consider loss to follow-up due to mortality in the analysis.

For the first research objective, concerning the effect of the programme on continuity in general practice, we compared the three coefficients of continuity of care between the cohorts. For the second research objective, concerning the effect of continuity on hospitalizations, we compared the cohorts with respect to rates of hospitalization, rehospitalization, and avoidable hospitalization and included the continuity of care coefficients as potential predictors (three separate models for each continuity coefficient). In this way, we examined the independent effect of continuity on hospitalizations (i.e. independent of receiving the programme).

We assessed models’ “goodness of fit” by the quasi-likelihood under the independence model criterion [20] and models with the best goodness of fit were preferred. We considered P-values smaller than 0.01 significant. Data storage and extraction was performed with MySQL Community Server × 64 (Oracle Corporation, Redwood Shores, CA, USA). Statistical analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, North Carolina, USA).

Patients in the strong general practice care programme had higher continuity in general practice than a comparable cohort of control patients. This means that they more often attended the same physician and had fewer handoffs. High continuity in general practice was associated with lowered hospitalization rates, independent of receiving the programme. Given the cross-sectional nature of the study, we cannot conclude that high continuity in general practice caused lowered hospitalization rates, even though this seems a plausible explanation.

The strengths of the study are the naturalistic setting, the high validity of the claims data, and the large-scale and sustained programme of strong general practice care. The health insurance claims data cover almost half of the population in the state, but specific groups (e.g. people with high income) are less well represented. Patients in Germany do not necessarily have a regular GP, because they have free choice of physician. In the broader evaluation of the programme for strong general practice care, it was decided to compose a control cohort of patients who had at least some continuity of physician. This might have led to an overestimation of the overall continuity of care in the control cohort and an underestimation of the beneficial effects of the programme, as medical care in the control cohort had higher continuity than average. Thus, in reality the effect of the programme on continuity in general practice may be larger than found in our study. Given the methods used for linking patients to physicians, we are fully confident that patients in the intervention cohort have been linked to a specific GP (regardless of whether this was the most frequently attended physician). The programme for strengthening general practice was only open for GPs and enrolment required an active decision, including written informed consent. In the control cohort, the method guaranteed that patients were linked to a GP as the medical discipline was recorded for all physicians and only GPs were included in the analysis.

As a background for understanding the findings of the study, some aspects of German healthcare need to be highlighted. As compared to countries that dominate the international research literature on primary care (e.g. United Kingdom, Denmark, Netherlands), the number of visits of patients to GPs is high in Germany. This is probably related to the reimbursement system, which relates payment to patient attendance per quarter of year and reimburses specific services separately. Furthermore, the number of hospital beds and the probability of hospitalisation is higher in Germany than in many other countries. This may provide an easy starting point for reduction of hospitalization.

Looking at other studies that used similar measures of relational continuity in general practice, we can conclude that the continuity in general practice is high in the observed cohorts. For instance, a study of general practice in England showed an average Usual Provider Continuity of 0.61 overall, albeit a little higher (0.70) in small practices with up to three physicians [3]. This index was 0.87 in the control cohort in our study, and higher in the strong general practice care programme. A study of general practice care in the United States showed a Usual Provider Continuity of about 0.80 on average [4], which is closer to our study but still lower. For a good interpretation of the findings of this study, it should be noted that the average number of visits of a patient to a GP is estimated on 10 to 15 per year, which is substantially higher than in many other countries.

Regardless of the absolute degree of the continuity of care, its lowering effect on hospitalization rates was found across the range of observed values across studies (e.g. Usual Provider Continuity values from 0.50 to 1.00). The effect of continuity on hospitalization in our study seemed even higher than in other studies. The English study looked at avoidable hospital admissions and found 3.87% lower rates in the tertile of high continuity as compared to the tertile of medium continuity [3]. The effect is (about three times) smaller than the effect in our study (as the relative risk reductions in Tables 3 and 4 refer to change of one decile on the continuity of care measure). The U.S. study reported about 7% lowering of total hospitalization rates, related to quintiles on the continuity of care measure [4]. This effect is (almost three times) smaller than the effect found in our study.

We can only speculate about the reasons for the stronger association between continuity of general practice and hospitalization rates in Germany, which is particularly striking as the absolute degree of continuity of care was already very high. The programme had a number of components that may have contributed to the lowering of hospitalization rates, such as the structured management of chronic diseases [11]. Also, the utilization of hospitals is relatively high in Germany compared to other countries. For instance, the Organisation for Economic Collaboration and Development Health Database reports 25.5 hospital discharges per 100 inhabitants in 2017 for Germany (around the 28.7 hospital admissions per 100 inhabitants in our study). The numbers of hospital discharges per 100 inhabitants are, for instance, 18.6 for France, 17.1 for Switzerland, 12.3 for the United Kingdom, and 9.2 for The Netherlands [21]. Although these numbers do not convey information on the appropriateness of hospital care, they may suggest a potential to reduce the number of hospital admissions in Germany. We assumed that lowered rates of hospitalizations are beneficial, not undesired delay of hospital admission, because of the wide availability and easy access to hospital care in most of South Germany.

In Germany, patients in nursing homes are medically managed by primary care physicians. If all other things are equal, we found that patients in nursing homes have a lowered probability for hospitalisation. It may be possible that the treatment and care for these patients is, on average, better than for patients at home. However, the number of patients in nursing homes was low, so we do not want to interpret this finding too prominently.

Future research should sort out the causality of the identified associations. For instance, the findings might be explained by a reversed causal explanation: lowered continuity in general practice (i.e. seeing more different physicians) after hospital admissions. Future analyses should model the timing of hospitalizations in relation to the experienced degree of continuity in general practice. This would also help to provide insight into the role of continuity of care in the causal chain underlying the lowering impact of the programme for strengthening general practice on hospitalizations. Randomized trials provide the strongest evidence for causality, but these may not be feasible for large-scale study in a naturalistic setting.

Future studies may explore whether suboptimal interpersonal continuity of care can be compensated for by information continuity and team continuity, which seems unavoidable in the context of current physician shortages and physicians’ perceived needs for work-life balance. In Germany, many general practice care practices are run by one or two physicians, although there is a trend towards practices with more staff (both physicians and assistants) [22]. This trend is present in other countries, such as the United Kingdom where continuity of general practice is lower than in Germany, particularly in large general practice care practices [3]. While this trend may be unavoidable, its potential impact on the quality and outcomes of healthcare needs attention.