Objective To assess the association between continuity of provider-adjusted regularity of general practitioner (GP) contact and unplanned diabetes-related hospitalisation or emergency department (ED) presentation.
Design Cross-sectional study.
Setting Individual-level linked self-report and administrative health service data from New South Wales, Australia.
Participants 27 409 survey respondents aged ≥45 years with a prior history of diabetes and at least three GP contacts between 1 July 2009 and 30 June 2015.
Main outcome measures Unplanned diabetes-related hospitalisations or ED presentations, associated costs and bed days.
Results Twenty-one per cent of respondents had an unplanned diabetes-related hospitalisation or ED presentation. Increasing regularity of GP contact was associated with a lower probability of hospitalisation or ED presentation (19.9% for highest quintile, 23.5% for the lowest quintile). Conditional on having an event, there was a small decrease in the number of hospitalisations or ED presentations for the low (−6%) and moderate regularity quintiles (−8%), a reduction in bed days (ranging from −30 to −44%) and a reduction in average cost of between −23% and −41%, all relative to the lowest quintile. When probability of diabetes-related hospitalisation or ED presentation was included, only the inverse association with cost remained significant (mean of $A3798 to $A6350 less per individual, compared with the lowest regularity quintile). Importantly, continuity of provider did not significantly modify the effect of GP regularity for any outcome.
Conclusions Higher regularity of GP contact—that is more evenly dispersed, not necessarily more frequent care—has the potential to reduce secondary healthcare costs and, conditional on having an event, the time spent in hospital, irrespective of continuity of provider. These findings argue for the advocacy of regular care, as distinct from solely continuity of provider, when designing policy and financial incentives for GP-led primary care.
- primary health care
- chronic disease
- continuity of care
- health policy
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
Strengths and limitations of this study
This study is the first to assess the impact of regularity of general practitioner (GP) contact adjusting for continuity of (GP) provider.
This study used two-part and zero inflated negative binomial models which allowed associations to be expressed conditional on having an event and incorporating the probability of having an event.
The assessment of the impact of provider-adjusted regularity of GP contact across subpopulations provides information that will facilitate more targeted approaches to primary care policy development.
This study included a comprehensive range of self-reported data on health and socioeconomic status not often available in linked data studies in addition to administrative data on current and previous health service utilisation.
This was a cross-sectional observational study over a 6-year period which limits the ability to infer causality. However, the inclusion of both health service use and outcomes prior to the study period in the modelling allowed us to partially ameliorate this limitation.
In line with other developed countries, chronic diseases predominate in Australia accounting for 66% of disease burden in 2011 and 73% of deaths in 2013.1 2 One strategy to address this challenge is to shift care from hospital to primary care. This is because primary care treatments are generally less costly3 and hospitals are not well-suited to manage chronic conditions.4 This approach is consistent with a body of literature on Ambulatory Care Sensitive Conditions; conditions for which hospitalisation may be avoidable if effective outpatient treatment is provided.5 In recognition of this developing evidence base, a National Chronic Disease Strategy was devised in Australia in 2005, which promoted the integration and continuity of care.6 Similar initiatives have been introduced in the United Kingdom7 and the USA.8
Although chronic disease management in primary care has the potential to be less costly and more appropriate for patients, primary care services in Australia have historically been episodic and reactive. Such models of care are likely inadequate for patients with chronic conditions and multimorbidities. Evidence to date suggests integrated care models with a focus on chronic disease management are important for increased health system efficiency and are more responsive to consumers’ needs and preferences.9 10 In countries such as Australia, where primary care operates on a fee-for-service basis, incentive payments can be used as a mechanism to modify primary care provider behaviours in terms of how healthcare is accessed and delivered.11 Improving planned chronic disease management by general practitioners (GPs), rather than focusing on provider continuity may be appropriate in health systems with diverse funding arrangements, as there is evidence from the United Kingdom12 13 and the USA14 that continuity of provider is falling.
Better linkage of care, with the GP as the central care provider, is underpinned by the theoretical notion that better continuity of care improves health and health system outcomes. GP provided care is publicly subsidised in Australia through the universal healthcare access programme, Medicare. GPs also provide a gateway to specialist and some allied health services, as a referral is required for publicly subsidised specialist or allied healthcare. The definition of ‘continuity’ is inconsistent and the mechanisms by which different components of continuity with a GP reduces hospitalisation are not well described. Most continuity indices focus on continuity of provider.15 16 One measure of continuity of provider is the usual provider concentration index (UPC) which is a simple measure of the proportion of visits to the same provider.17 Under most previous measures, a patient seeing a single provider on a regular, frequent basis may have a similar continuity score to a patient seeing a single provider on a very fragmented basis. For this reason, a body of work from Australia has evolved to explore the concept of ‘regularity’ of GP contact.18–20
Regularity refers not to the number of GP visits, but rather measures the dispersion of GP visits over time, with more even dispersion indicating better regularity. It has been shown previously that the use of financial incentives to GPs increases the regularity of primary care contact in the following year without increasing frequency of contact.19 More recently, we have reported decreased rates and costs of diabetes-related hospitalisations with increasing GP regularity.20 This finding suggests that regularity is potentially suitable as a target for policy intervention and provides some justification for the interpretation of regular primary care as indicative of proactive management of a patient or condition, as opposed to reactive or unplanned care. An important limitation of previous work has been the inability to adjust for continuity of provider when evaluating the impact of regularity. No attempt at disentangling the impact of regularity from that of provider continuity has been reported to date. Thus, despite the relevance for policy development, whether increased regularity is actually a proxy for, or a consequence of, increasing continuity of provider, or is a discrete facet of continuity of care is unknown.
The aim of this study was to assess the continuity of provider-adjusted association between regularity of GP contact and unplanned diabetes-related hospitalisations and emergency department (ED) presentations.
This was an observational cross-sectional study using self-reported survey data linked with routinely collected unit record administrative health data from 1 July 2009 to 30 June 2015. Reporting follows the Reporting of studies Conducted using Observational Routinely collected health Data (RECORD) guidelines.21
Patient and public involvement
A consumer representative was involved in the design of the grant used to fund this research. The 45 and Up Study, which provided data for this project, maintains a repository of published research using this cohort online.
The study used both self-reported and routinely collected administrative data linked at the person level from the 45 and Up Study.22
The Sax Institute’s 45 and Up Study is a longitudinal cohort study of 2 66 885 participants, aged 45 years and above in the state of New South Wales (NSW), Australia. Prospective participants were randomly sampled from the Australian Government Department of Human Services (DHS), formerly Medicare Australia, enrolment database and recruited from 2006 to 2009. The study methods are described in detail elsewhere.22 Briefly, participants completed a baseline health and lifestyle questionnaire and consented to follow-up and linkage to routine health databases. The overall response rate was 18%.22
The data sources linked and used in this study included: (1) the 45 and Up Study baseline questionnaire (https://www.saxinstitute.org.au/our-work/45-up-study/); (2) the NSW Admitted Patient Data Collection (APDC) which provided all hospital separations in public and private hospitals in NSW (2005–2015); (3) the NSW ED Data Collection (EDDC) which provided information on all ED presentations from 2006 to 2015; (4) the Pharmaceutical Benefits Scheme (PBS) which provided information on subsidised prescription medicines dispensed (2005–2015); (5) the Medical Benefits Schedule (MBS) which provided records for all claims for medical and diagnostic services provided through Medicare, Australia’s universal health insurance scheme (2005–2015) and (6) the NSW Register of Births Deaths and Marriages (RBDM) (2006–2015). The linkage of APDC, EDDC and RBDM to the survey data was conducted by the NSW Centre for Health Record Linkage. MBS and PBS data were linked by the Sax Institute using a unique identifier provided by the DHS. Quality assurance data on the data linkage show false-positive and false-negative rates of <0.5 and <0.1%, respectively.23
Ascertainment of previous history of diabetes and entry into the study cohort
This study used a cohort of 45 and Up Study participants with a history of diabetes mellitus (excluding malnutrition-related diabetes mellitus, International Statistical Classification of Diseases, 10th Revision, Australian Modification (ICD-10-AM) code E12) on or before 31 December 2009. This was ascertained via: (1) self-report from the baseline questionnaire using the question ‘Has a doctor ever told you that you have diabetes?’; (2) evidence of hospitalisation indicating diabetes from 2005 to 2009 using the ICD-10-AM codes E10, E11, E13, E1424 or (3) a PBS claim indicating a dispensing from 2005 to 2009 using the following Anatomical Therapeutic Chemical codes A10A (insulins and analogues) and A10B (blood glucose lowering drugs excluding insulins).
Participant inclusion and exclusion criteria
We included participants with diabetes who had at least three GP contacts (minimum required for the UPC metric) during the study period. We excluded participants who died before 30 June 2011 to allow for a minimum of 2 years of follow-up (figure 1). Exposure (ie, patterns of GP contact) and outcomes were ascertained from 1 July 2009 to 30 June 2015 or, for participants who died during the study period, until the end of the financial year (ie, 30 June) prior to death. Over the follow-up period, a small but unknown number of participants may have moved out of NSW, where no data were collected. This would then affect collection of outcome and exposure data beyond the date in which the individual left NSW. Among those continuing to reside in NSW, follow-up for hospitalisation is considered to be ~98%.25
Study outcomes were the number, associated bed days (for hospitalisations) and cost of unplanned diabetes-related hospitalisations or ED presentations occurring during the study follow-up period.
Hospitalisations included those classified as diabetes-related potentially preventable hospitalisations by the National Health Performance Framework (E10, E11, E13, E14)24 or where diabetes was identified as a significant risk factor by Davis et al.26 The ICD codes listed by Davis et al were those where diabetes was considered to increase the risk of hospitalisation for that condition. The list is freely available on table 1 of these authors’ publication.26 Diabetes related, rather than all hospitalisations were chosen as an outcome because primary care for a cohort defined by having diabetes could not be reasonably expected to be associated with lower all-cause hospitalisation, and thus any association identified would be outside of the study aims. Hospitalisations were further limited to those identified as unplanned, identified using the Emergency status (urgency of admission) variable in the APDC, as these are most likely the hospitalisations that are most amenable to the potentially protective effect of better continuity of primary care. Same-day hospitalisations for routine dialysis were excluded.
Supplementary material 2
Unplanned diabetes-related ED presentations were ascertained from EDDC data using either the ICD diagnosis or the Systematized Nomenclature of Medicine—Clinical Terms (SNOMED-CT) code, depending on availability in the data. ICD codes used to capture diabetes were those detailed above mapped to the ICD version used on the EDDC record. SNOMED-CT codes included those with a description relating to diabetes excluding where the wording included ‘not related to diabetes’ or relating to diabetes insipidus.
Interhospital transfers, as well as ED admissions resulting in hospitalisation, were counted as a single episode of care, with bed days calculated for each single episode of care.
Hospitalisation costs were assigned based on average public hospital Australian Refined Diagnostic-Related Group (AR-DRG) costs reported in the National Hospital Cost Data Collection round 18 (2013/2014) as this was the most recent year of DRG costing in the APDC data.27 Costs were applied based on the AR-DRG code recorded for each admission. Cost of ED presentations was determined using cost weight for each urgency disposition group (UDG) using the National Efficient Price Weight Determination 2018–2019 Price Weight table.28 UDGs classify patients into 12 groups based on the type of visit, episode end status and triage code.29 UDGs were generated to reflect the UDGs reported on the National Hospital Cost Data Collection on the basis of the triage category,1–5 mode of separation and visit type reported in the EDDC data. All costs were inflated to 2017 Australian dollars using the Consumer Price Indices and included costs of nested events, where present.30
GP contact was captured via MBS claims for ‘Attendances by General Practitioners’.31 We used a modified version of the regularity index, described previously.19 Both indices use the number of days between GP visits within an ascertainment period. Whereas the original regularity score (RV) was constructed using the formula RV=1/(1+variance (days)), the modified regularity index was calculated using the coefficient of variation (Cv) in place of the variance. Cv was calculated with the formula:
The modified index produces a unitless measure of variation, which is less correlated with frequency compared with previous measures.20 The modified regularity index (RCv) was constructed using the formula RCv=1/(1+Cv (days)). This resulted in a score between 0 and 1 per individual, with 1 indicating perfectly regular contact. This score was separated into quintiles from least to most regular, using the range of scores observed in the cohort.
In addition to regularity we also used UPC, which calculates the proportion of all GP visits made to the most frequently seen GP over the study period and ranges from 0 to 1.15 32 UPC was measured over the study period using deidentified provider numbers in the MBS data.
The 45 and Up Study questionnaire data included self-report information on key potential confounders and mediating factors including: baseline age; sex; marital status; born in Australia; Indigenous status; current housing; household income; education level; smoking history, intensity and age stopping; alcohol use; physical activity33; time spent sitting; body mass index; psychological distress34; level of limitation reported; social support35; socioeconomic status36; and Accessibility/Remoteness Index of Australia.37 Covariate categories are provided in online supplementary file 1.
Supplementary material 1
Comorbidity was ascertained using the Multipurpose Australian Comorbidity Scoring System,38 defined as the sum of comorbidities excluding diabetes at 1 and 5 years prior to study entry. We also used the PBS data to calculate the Rx-Risk index (ie, number of condition groups excluding diabetes for which medicines were dispensed) at 1 and 5 years prior to entry to the study period.39
Health service use during and prior to the study period was captured in several ways. The number of GP and specialist physician visits 4 years before and during the study period was ascertained separately using MBS data. The frequency of contact was captured as: (1) a count of the number of days each person had a GP or specialist physician contact and (2) the SD of the annual count of days with a contact during the ascertainment period. Variables also captured if participants had a PBS records for any diabetes-related medicines, hospitalisations for dialysis, MBS diabetes cycle of care claim40 and other chronic disease related MBS-funded primary care services or glycosylated haemoglobin (HbA1c) testing during or in the 4 years prior to the study period. Finally, the number of unplanned diabetes-related hospitalisations and ED presentations in the 3 years prior to study entry was also included as a potential confounding factor.
To reduce ascertainment bias, the method (ie, self-report only, APDC only, PBS only or combinations of the three) and first year the individual who identified as having diabetes was entered into the model. In addition, a binary variable was used to determine if the participant died during the 6 years of the study period. Person-time at risk of the outcome event was included in the count models.
Analyses were conducted using Stata SE Version 15 (Stata, College Station, Texas, USA). The relationship between regularity, UPC and count outcomes was evaluated in two stages.41 First, a logit model was fitted for the probability of observing the relevant outcome. Second, a negative binomial model, conditional on a positive outcome was developed. The number of days out of hospital (conditional on being alive) was used as the time at risk variable for the number of unplanned diabetes-related hospitalisations and ED presentations, while the number of years in the study cohort under observation was used as the exposure variable in the evaluation of number of bed days for unplanned diabetes-related hospitalisations. The overall relationship, accounting for the probability of having an outcome, was subsequently evaluated using zero-inflated negative binomial (ZINB) multivariable regression, which included separate components to model zero and non-zero outcomes to account for overdispersion of the data and the high proportion of person-time with no outcome. The previously defined logit and negative binomial models were used for each part of the ZINB model. For the cost outcome, a two-part model (Stata -twopm-) incorporating a logit (first part) and generalised linear model (GLM) with a gamma family and a cubed root link function for the second part were used. The GLM family was determined using the Modified Park Test. The most appropriate link function, conditional on the gamma family, was then evaluated using power links in 0.1 intervals between 1 (identity) and 0 (log) using Akaike Information Criterion (AIC)/Bayesian Information Criterion (BIC) and the Pearson correlation test. Once the best link function was determined, the specification of the model was evaluated using the link test.
Since the model used for cost (Stata -twopm- module) is a wrapper that facilitates computation of the combined marginal effects and adjusted predictions from the two separate models (logit and conditional GLM), comparative coefficients (ie, inflated for the probability of a zero outcome) across the outcomes do not exist. Further, the coefficients from the ZINB models cannot be directly interpreted as the impact of regularity adjusting for the probability of having an event. Rather, average marginal effects (AMEs) need to be estimated. Thus, to facilitate comparison of the impact of regularity and UPC, considering the probability of having an event in this population across outcomes, the AMEs, rather than the coefficients, were reported.
Models were built using backwards selection with retention of covariates in the parsimonious model based on statistical significance (p≤0.05) or a change in the effect size in respect to regularity of ≥10%. Regularity and UPC were retained in all models a priori. Competing models were evaluated using the AIC and BIC. All models incorporated robust SEs.
Nine per cent (23 926) of 45 and Up Study participants self-reported a diagnosis of diabetes (figure 1). Of those not self-reporting a diagnosis of diabetes, 3373 had a diabetes-related hospitalisation prior to 2010 and 3038 of the remaining 239 586 had a diabetes medicine dispensed. The resulting diabetes cohort of 30 337 was reduced to 27 409 due to potential linkage error (n=182), dying prior to 1 July 2011 (n=2264) or having less than three GP visits during the study period (n=482).
Cohort characteristics and unadjusted outcomes are summarised in table 1, with full characteristics defined by regularity quintile and UPC category in online supplementary file 2. At entry, the diabetes cohort was 46% female. The mean age was 68 years and 1.4% of the cohort was Indigenous. The majority (71%) were born in Australia and 73% were obese or overweight. Just over half were never smokers (51%), had a tertiary qualification (55%) and lived in a ‘highly accessible’ location (53%). The cohort was followed for a mean of 5.7 years and averaged 13 GP visits annually. The mean reported days with a GP visit throughout the 6-year study period was higher for the lowest quintile of regularity (69) than the highest quintile of regularity (66). The number of diabetes ‘cycle of care’ MBS claims, HbA1c tests and chronic disease management plans were higher for the highest quintile of regularity than for the lowest. Eighteen per cent died during follow-up, with 26% of the deaths coming from the lowest quintile of regularity. Twenty-one per cent of the diabetes cohort had a diabetes-related hospitalisation or ED presentation during the study period. Forty-three per cent of participants had a UPC between 0.76 and 1.00 (online supplementary file 2). There was a positive crude association between increasing UPC and increasing regularity (table 1).
Outcomes of conditional models
Table 2a shows the coefficients for regularity and UPC from the multivariable logit model of unplanned diabetes-related hospitalisation or ED presentation. The full model output is shown in online supplementary file 3. A reduction in the probability of unplanned diabetes-related hospitalisation or ED presentation was observed for each subsequent quintile compared with the lowest quintile ranging from −0.13 to −0.28. Relative changes in the probability of an unplanned diabetes-related hospitalisation alone were similar (online supplementary file 3). Increasing UPC did not significantly change the probability of an unplanned diabetes-related hospitalisation or ED presentation (−0.03, 95% CI −0.19 to 0.13). The adjusted probability of an unplanned diabetes-related hospitalisation or ED presentation over a 6-year period was on average 0.23 for individuals in the lowest quintile of regularity and reducing to 0.20 for those in the highest regularity quintile (figure 2A). UPC did not modify the association between quintile of regularity and the predicted probability of having a diabetes-related hospitalisation or ED presentation (online supplementary file 4a).
Supplementary material 3
Supplementary material 4
Table 2b–d shows the model outputs for study outcomes, conditional on having one or more events over the study period; the full model outputs are shown in online supplementary file 5. There was no consistent association between increasing regularity and the number of diabetes-related hospitalisations or ED presentations. The relative number of unplanned diabetes-related hospitalisations was significantly reduced for those with low (−6%, 95% CI −13% to 0%) and moderate (−8%, 95% CI −15% to −2%) compared with those with the lowest regularity. However, no significant difference existed between individuals in the highest two quintiles of regularity and their counterparts who had the lowest regularity. UPC was associated with a decrease in the relative number of diabetes-related hospitalisations or ED presentations (−11%, 95% CI −21% to −1%). Predicted diabetes-related hospitalisations or ED presentations per 100 person-years at risk ranged from ~38 in the lowest and high quintiles of regularity to 35 in the moderate quintile of regularity (figure 2B). Increasing UPC was independently associated with a reduction in the adjusted rate of diabetes-related hospitalisations or ED presentations, but the association was not differential by quintile of regularity (online supplementary file 4b).
Supplementary material 5
There was a clear negative association in the relative number of bed days, conditional on an event occurring; −30% (CI −38% to −22%) for low to −44% (95% CI −53% to −36%) for high quintile compared with the lowest regularity quintile (table 2c). Conversely, UPC was also positively associated with the relative number of bed days (17%, 95% CI 4% to 30%). The adjusted predicted rate of bed days decreased by quintile of regularity, with a predicted 561 bed days per 100 person-years at risk for the lowest quintile of regularity, and 373 bed days for the highest quintile of regularity (figure 2C). No differential effect of quintile of regularity by UPC score was observed (online supplementary file 4c).
A similar trend as described for bed days was observed for cost, with all coefficients negative relative to the lowest regularity quintile (table 2d). The reduction in cost relative to the lowest quintile was greatest in the high quintile (−41%, 95% CI −56% to −27%), with a slightly lower reduction observed in the highest quintile (−39%, 95% CI −55% to −23%). The adjusted predicted annual cost per 100 person-years at risk shows a range from $A1 154 557 for the lowest to $A784 507 for the high quintile of regularity. No significant association with UPC was observed (13% increase, 95% CI −9% to 36%). Online supplementary file 4d shows a non-differential effect of UPC by quintile of regularity.
Outcomes of unconditional models
There was no significant difference in the number of diabetes-related hospitalisations or ED presentations across quintiles of regularity, when accounting for the probability of having an event (table 3). Similarly, the association of regularity on bed days for unplanned diabetes-related hospitalisation became non-significant. In comparison, there was a significant negative association between regularity and cost across all regularity quintiles but as described below the magnitude of the association was substantially reduced from that observed in the conditional model. The AMEs of regularity in the conditional cost model range from −$A12 920 (95% CI −$A21 2800 to −$A4560) for the low regularity quintile to −$A22 200 (95% CI −$A30 290 to −$A14 120) for the high regularity quintile. However, the AMEs for the same quintiles were reduced by approximately 70% when the probability of having a hospitalisation/ED presentation was incorporated into the model: −$A3798 (95% CI −$A5831 to −$A1765) for low regularity to −$A6530 (95% CI −$A8505 to −$A4556) for high regularity. The significant positive AMEs with increasing UPC with respect to bed days and cost in the conditional models disappeared in the unconditional models.
The probability of having an unplanned diabetes-related hospitalisation or ED presentation was inversely associated with regularity of GP contact, after adjusting for continuity of provider. Conditional on having an event, increasing regularity was associated with both a lower cost and reduced days in hospital. The association with cost, while somewhat attenuated, persisted when the probability of having an event was taken into account. However, the association was not significant with respect to bed days.
These findings lend support to primary care policies where proactive managed care by any GP is incentivised. In Australia, there is no formal requirement to register at a GP practice, whereas in the United Kingdom, people need to register with the National Health Service at a single GP practice.42 Campbell et al 12 43 report a fall in continuity of provider following implementation of the Quality Outcomes Framework (QOF), a practice-level financial incentive to improve clinical management of chronic disease in the United Kingdom. Our results suggest that this might not matter in terms of achieving a reduction in hospitalisations for people with diabetes, since the effect for regularity was larger than those of UPC when modelled together. This suggestion is corroborated by a slight fall in emergency hospitalisations for practices with more diabetic patients with moderate, as compared with poor glycaemic control, in England following implementation of the QOF.44 While there is no QOF in Australia, initiatives such as the MBS diabetes ‘cycle of care’ payment are similarly aimed at improving the quality of chronic disease management.40 Also using 45 and Up Study survey data, Comino et al 45 have reported a lower adjusted rate of hospitalisation for people with diabetes for whom a diabetes cycle of care was prepared, or a GP management plan reviewed. Diabetes complications vary in both number and severity.46 The association between increasing regularity and fewer bed days (in the unconditional model) and lower cost (in both models) may indicate a lower severity of complications on average at hospital admission. Increasing regularity of GP contact will not completely negate unplanned hospitalisations, though timely intervention may allow hospital-level care to be initiated earlier. For example, a diabetic foot ulcer may still require hospitalisation for management, but avoid progression necessitating more severe and costly management.47 Ballo et al 48 in a study assessing chronic care management for people with heart failure found an increase in hospitalisation among this group compared with controls. This was primarily driven by planned hospitalisations, for which there was a greater increase relative to unplanned. Once in place, removal of quality incentives can also change clinician behaviour, as shown recently with a decrease in documented quality measures following cessation of some of the UK’s QOF elements.49
For the cost outcome, the AME was −$A6530 for the high quintile of regularity yielding a total cohort hospitalisation cost reduction of $A179 million. The cost-effectiveness of increasing regularity is out of scope of this study; however, this is a potential avenue for further study. Another important issue to consider is whether modelled outcome changes by regularity quintile are due to inherent differences in the people occupying each quintile, with these differences rather than any effect from GP care explaining the observed differences. There were some demographic differences across quintiles as indicated in table 1. In our study, a wide range of covariates were accessible including sociodemographic, psychosocial, health behaviours and prior health service use, and while we cannot rule out residual confounding, the adjustment for a wide range of covariates that could plausibly affect the association reduces the risk of substantial confounding affecting interpretation.
Previous work has demonstrated an association between managed care and hospitalisations.18 50 Barker et al 51 have reported an association between increasing UPC and fewer ambulatory care sensitive condition hospital admissions, with a 6.2% decrease per 0.2 increase in UPC. Admissions for 22 conditions were considered by these authors making the study cohort more heterogeneous than the single-condition cohort in this study. These authors adjusted for GP frequency, but not regularity. While determination of the impact of continuity of provider was not the aim of our paper, we found a negative association between UPC (adjusted for regularity of contact) and the days spent in hospital and cost of unplanned diabetes-related hospitalisations or ED presentations, conditional on an event occurring. However, a negative association with the number of unplanned diabetes-related hospitalisations or ED presentations was not observed. Our results are, therefore, discordant with literature reporting a negative association between continuity of provider and hospitalisation.51–57 This may be explained by our simultaneous adjustment for regularity and continuity of provider. If increased continuity of provider improves regularity of contact, then studies reporting UPC without adjusting for regularity may have not differentiated between these two facets of continuity of care.
This study had several strengths. Adjustment for both regularity and continuity of provider shows that the association between regularity and study outcomes is not solely due to frequency or continuity of provider. The availability and inclusion of covariates from the 45 and Up Study and administrative data have also helped to reduce the risk of associations being due to different characteristics in each regularity quintile. Effect modification was investigated by testing the significance of an interaction term and reporting the association with regularity across the levels of UPC. Our definition of the outcome as being unplanned diabetic-related hospitalisation or ED presentations has also facilitated a deeper analysis of the impact of GP contact on a subset of diabetic-related hospitalisations that are more likely to be amenable to change by high-quality primary care. The two-part analytical methods allowed results to be presented both conditionally and unconditionally. This allowed assessment of the association between regularity and the probability of hospitalisation, the impact on additional hospitalisation for those previously hospitalised and overall for the population with diabetes. The latter is important because primary care policies are usually implemented across populations, and thus determining the value (impact vs cost) of the policy overall is important. Identifying if the impact differs across subpopulations is also informative, as this knowledge may facilitate more targeted approaches to policy development. There were also limitations to this study. This was a cross-sectional observational study over a 6-year period. In this study, we have reported associations, rather than to make any causal inferences. Temporal ordering of exposure followed by outcome is not given because of the cross-sectional study design. The lack of detail (eg, of length of GP consults or clinical data to indicate disease severity) in the analysis also reduces the level of detail analysed regarding these interactions. Migration outside of NSW leads to loss to follow-up. Based on mean total international and interstate migration for NSW between 2009/2010 and 2014/2015 (financial years), this proportion is at most 4.8%.58–60 However, 81.3% of these people were aged below 44 years and thus, while the total proportion of study participants affected is unknown, the impact on results is likely to be minimal (<1% of cohort). The results of this study are not generalisable to people with diabetes below the age of 45 years. Though the response fraction in the 45 and Up Study is 18%, the generalisability of results has been assessed by comparing effect measures for this cohort with those from the NSW Population Health Survey, which had a ~60% response.61 These authors concluded that their ‘findings show that broad ranges of exposure-outcome relationships estimated from two studies of the same population remained consistent regardless of the underlying response rate’.61
This is the first study to assess both regularity and continuity of provider in primary care, demonstrating an independent association between increasing GP regularity and the probability of an unplanned diabetes-related hospitalisation or ED presentation, bed days and associated costs. Based on these results, continuity of provider and regularity should be considered as distinct phenomena in future analyses, since this would aid in determining which facets of continuity are driving outcomes and aid in decisions regarding promotion of incentives that support (1) ongoing contact with the same provider (without a focus on regularity of care), (2) regular care (by any provider) or (3) regular care by the same provider. Future research should incorporate a measure of regularity, separate from provider continuity, to enable further assessment of systems-level policies to improve chronic disease management in settings with a range of funding models in primary care. The findings in this study argue for chronic disease management incentives which are not necessarily tied to a provider and invites similar adjustment in future analyses assessing the impact of GP-led primary care financial incentives. Given falling continuity of provider in some settings, levers to improve coordination of care irrespective of provider may help to reduce secondary care requirements for chronic conditions managed in primary care.
This research was completed using data collected through the 45 and Up Study (www.saxinstitute.org.au). The 45 and Up Study is managed by the Sax Institute in collaboration with major partner Cancer Council NSW, and partners: the National Heart Foundation of Australia (NSW Division), NSW Ministry of Health, NSW Government Family & Community Services – Ageing, Carers and the Disability Council NSW, and the Australian Red Cross Blood Service. We thank the many thousands of people participating in the 45 and Up Study. We also acknowledge the Commonwealth Department of Human Services for the provision of MBS and PBS data.
Contributors REM developed the study concept, conducted the analyses, interpreted results and drafted the manuscript. CMW contributed to the study design, interpreted results and drafted the manuscript. DY contributed to the study design and reviewed the draft manuscript for important intellectual content. DBP and MH reviewed the manuscript draft and provided comments on important intellectual content. All authors reviewed the manuscript and approved it for submission.
Funding This study was part of a project funded by the Australian National Health and Medical Research Council (APP1078345). The study funders had no role in the study design, conduct, manuscript writing or decision to submit for publication. Due to data access restrictions placed on the 45 and Up Study data, only the approved analysts (REM and DY) had access to the data for analysis, while all remaining authors, external and internal, had full access to all statistical reports and tables. REM and DY can take responsibility for the integrity of the data, while all authors can take responsibility for the accuracy of the data analysis.
Competing interests None declared.
Ethics approval Institutional ethics committee approval was obtained from Curtin University Human Research Ethics Committee (RD-42-14) and the NSW Population and Health Services Research Ethics Committee (HREC/17/CIPHS/37). Consent was given by all participants in the 45 and Up Study for their information to be used in approved studies, and for follow-up and data linkage. The conduct of the 45 and Up Study was approved by the University of NSW Human Research Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The data that support the findings of this study are available from the relevant data custodians of the study data sets. The 45 and Up Study data were used under licence for the current study. Restrictions by the data custodians mean that the data are not publicly available or able to be provided by the authors. Researchers wishing to access the data sets used in this study should refer to the 45 and Up Study application process (www.saxinstitute.org.au/for-researchers) and the Centre for Health Record Linkage application process (www.cherel.org.au/apply-for-linked-data).
Patient consent for publication Obtained.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.