Impact of a Patient-centered Medical Home on Healthcare Utilization for Patients With Complex Needs in Singapore: a Matched Cohort Study

Silvia Yu Hui Sim (  sim.silvia@gmail.com ) Geriatric Education & Research Institute https://orcid.org/0000-0001-5196-8271 Junxing Chay Duke-NUS Medical School https://orcid.org/0000-0002-3291-4523 Soon Hoe Ho Geriatric Education & Research Institute Mimaika Luluina Ginting Geriatric Education & Research Institute https://orcid.org/0000-0002-0319-8097 Zoe Zon Be Lim Geriatric Education & Research Institute https://orcid.org/0000-0002-8256-9924 Grace Sum Geriatric Education & Research Institute https://orcid.org/0000-0002-0309-2700 Joanne Yoong University of Southern California https://orcid.org/0000-0002-0162-9885 Chek Hooi Wong Tsao Foundation https://orcid.org/0000-0002-3048-0477

models of care is inadequate in managing these needs. , When multiple conditions are managed in silos, issues such as unnecessary polypharmacy, adverse drug interactions and treatment gaps tend to arise. 5 Fragmented care may also overlook non-medical in uences that add complexity to the management of these patients, including socio-economic challenges, family stressors and mental health issues such as depression. This adds burden to patients and caregivers by requiring them to navigate between multiple service providers and reconcile various treatment and services, resulting in patients with complex needs falling through the cracks and/or presenting to inappropriate healthcare settings. Studies have shown that patients with multiple chronic conditions are far more likely to incur hospital admissions for ambulatory care-sensitive conditions, and found a curvilinear, almost exponential relationship between multiple chronic conditions and healthcare utilization and costs.
Singapore, a small developed country in Asia, faces similar challenges to its health system with increasing numbers of patients with complex needs. It is estimated that 1 out of 5 Singaporeans will be aged 65 and above by year 2030. The ageing trend is accompanied by increasing chronic disease burden.
The proportion of Singaporeans with ≥3 chronic conditions rose from 19.8-37.6% within a decade, from 2009 to 2017. Over the same period, those with functional di culties in ≥1 basic activity of daily living (ADL) rose from 6.3-9.3%, and those with di culties in ≥3 instrumental activities of daily living (IADL) rose from 6.4-8.3%. Like most developed countries, Singapore's healthcare system was originally designed for the acute, episodic healthcare needs of a younger population of the past. 6 With the evolving demographics and healthcare needs, it is acknowledged that a stronger primary care model is required to provide more appropriate care.
The Patient-Centered Medical Home (PCMH), a primary care model, has been advocated as a more appropriate model for patients with complex needs. 8 Contrary to disease-speci c models of care, the PCMH model aims for comprehensive management of healthcare needs, recognizing the interactions between various medical conditions and non-medical challenges such as social and nancial needs. This is accomplished through improved communication and coordination between the patient, family and service providers. 8 The PCMH model has been shown to reduce inappropriate hospital-based healthcare utilization with improved management of care needs in the primary care setting and coordination of referrals to more appropriate care settings. , Most evaluation studies have found the PCMH model to be most effective in populations that are older or have higher needs, resulting in better health outcomes and reduced hospital-based healthcare utilization. , , , , , , In contrast, studies that found non-signi cant changes tend to be amongst populations with lower needs, or examine very early stages of PCMH implementation. 21, Moreover, implementation and evaluation of PCMH models have been carried out predominantly in the West, while less is known about its effectiveness in an Asian context. This study aims to assess the impact of a PCMH demonstration on healthcare utilization for patients with complex needs in Singapore. This, to the authors' knowledge, is the rst study to evaluate the impact of PCMH on healthcare utilization in Singapore and in Asia. This study would address the existing knowledge gaps about the effectiveness of PCMH in an Asian context and contribute to the wider discussion on the delivery of primary care for adults with complex needs.

Study design
This is a matched cohort study using a difference-in-difference approach to analyse healthcare utilization of PCMH study participants.

Intervention
This study assessed the Community for Successful Ageing (ComSA)-PCMH, launched in November 2016 in a geographically de ned central region of Singapore. Adults aged 40 and above identi ed to have high biopsychosocial needs by a locally-validated risk screening tool or clinical judgment, and residing within the geographic area were recruited to the clinic. These included referrals from specialist clinics and primary care clinics from the public regional health system, other private primary care and community partners as well as patients who walked in without referrals.
ComSA-PCMH consists of a primary care clinic led by family doctors trained in geriatric primary care and care coordination, as well as a home-based care management service led by medical social workers and nurses to support adherence to care plan, care coordination to manage nancial problems, health behaviour challenges, family con ict, caregiver stress and social engagement. The initial visits comprise a comprehensive needs assessment and care plan development with patient and caregivers. This is followed by regular reviews and acute treatments if needed. ComSA-PCMH delivers care that is integrated between the primary care clinic, home-based care management, the referring acute hospital and other community-based services. Patients are empanelled within the integrated PCMH with a regular primary care physician and home care team for lifelong engagement, to facilitate continuity of care and sustained patient-provider relationship. Patients are only discharged if they had no contact with ComSA-PCMH for >18 months, became institutionalised in nursing homes, homebound or passed away.

Sample
This study recruited 184 patients within their initial visits to ComSA-PCMH, from October 2017 to April 2019. Controls were selected from a neighbouring geographical region with similar population-level demographics of age, ethnicity, housing type and disease prevalence, and is served by the same public regional health system that did not offer the PCMH model of care. As participants had complex medical and non-medical needs, the sampling frame for controls was derived from the administrative records of a wide range of healthcare services, namely ve public primary care clinics (termed 'polyclinics' in Singapore) and specialist outpatient clinics (SOC), emergency department (ED) and impatient services under the public regional health system, from 1 September 2014 to 31 March 2020. The sampling frame consisted of 11,269 unique persons residing in the control region. Persons who had passed away during the period of analysis or had incomplete data were excluded for a complete case analysis ( Figure 1).
Matched controls were selected using Coarsened Exact Matching (CEM) on variables that may affect healthcare utilization patterns: birth year, gender, housing type, ethnicity, weighted Charlson Comorbidity Index (CCI) and whether the person had any inpatient admission or specialist outpatient clinic visit in the quarter before enrolment. Variables were temporarily coarsened into substantively meaningful groups (Table 1) and all exact matches were identi ed from the control group. Original, un-coarsened values were retained for analysis. CEM was selected over Propensity Score Matching as it can achieve lower levels of covariate imbalance, model dependence and bias. This matching method yielded different proportions of PCMH study participants and controls across the different coarsened typologies. All matched units were included in the analysis and weighting was used to account for the unequal strata sizes. The nal sample consisted of n=165 PCMH study participants and n=5,385 controls ( Figure 1).

Data sources
The study extracted administrative records from a public regional health system. This included records of polyclinic visits, SOC visits, ED visits, inpatient admissions and demographic data. From these records, we calculated the unique number of visits per person, per calendar-quarter (i.e. person-quarter) and CCI for all PCMH study participants and controls.
For study participants who were not in the public regional health system's administrative records, we derived demographic data and CCI from the PCMH clinic administrative database.

Data analysis
Difference-in-differences was used to identify changes in healthcare utilization of PCMH study participants pre-and post-enrolment into PCMH, relative to matched controls. This design enables assessment of whether enrolment into PCMH was associated with changes in healthcare utilization over time that were statistically different from the secular trend.
Regression analysis was conducted by tting a two-part model for each outcome measure. For the rst part, a logistic regression model was tted where the dependent variable was whether there was any (i.e. ≥1) healthcare utilization event (binary yes/no) per person-quarter. For the second part, a Poisson regression was tted, where the dependent variable was the number of healthcare utilization events per study person-quarter, among those who had any healthcare utilization events. This was conducted to handle the high number of person-quarters with zero healthcare utilization events in the data. ,, The main independent variable assessed is time since enrolment into PCMH. This was modelled using dummy variables to identify any non-linear effects over time. We included individual controls for each quarter, up to 7 quarters pre-enrolment and 3 quarters post-enrolment. Effects for time periods 8 quarters or more pre-enrolment and 4 quarters or more post-enrolment were treated as constant and grouped. In this analysis, enrolment refers to the calendar quarter (3 months) in which the PCMH study participant was enrolled into PCMH. Baseline referred to the quarter before enrolment. Controls were never enrolled into PCMH and hence took zero values for all dummy variables indicating time since enrolment.
Potential confounders were controlled for in both parts of the model: birth year, gender, ethnicity, housing type, CCI and treatment group (PCMH, control) to account for any unobserved difference between groups. Calendar time was also adjusted for, to account for secular changes in healthcare utilization, modelled using dummy variables to identify any non-linear trends. The last calendar quarter in this analysis (i.e. 2020Q1) is of particular interest as COVID-19 had spread to Singapore during that period. An interaction term (2020 × PCMH) was included to account for any differential effects that PCMH study participants sustained in 2020Q1 compared to controls. Cluster-robust standard errors at person-level were used to account for potential within-person correlation of healthcare utilization over time.  At baseline, both PCMH study participants(P) and controls(C) had highest utilization in SOC (P:1.47visits/person-quarter; C:1.22visits/person-quarter), followed by polyclinics (P:0.61visits/person-quarter; C:0.88visits/person-quarter), ED (P:0.16visits/person-quarter; C:0.08visits/person-quarter) and inpatient admissions (P:0.11visits/person-quarter; C:0.05visits/person-quarter). PCMH study participants had signi cantly fewer polyclinic visits (t(177)=-3.25, p=0.001) and marginally more ED visits (t(165)=1.88, p=0.06) and inpatient admissions (t(165)=1.95, p=0.05) per person-quarter compared to controls. (Table 2). Relative increases in the healthcare utilization of PCMH study participants compared to controls were observed over the pre-enrolment quarters. Then, relative reductions were rst observed in polyclinic visits in the quarter before enrolment, followed by SOC visits two quarter after enrolment and ED visits three quarters after enrolment, continuing till the end of the analysis period.

Discussion
Reductions in the healthcare utilization assessed were observed after enrolment into PCMH. These effects were robust even after adjusting for potential confounders, secular trends and differential changes in healthcare utilization that PCMH study participants and controls might have during the rst quarter of Covid-19 outbreak in Singapore.
These ndings are aligned with previous evaluation studies that also found reductions in SOC visits, ED visits and inpatient admissions associated with PCMH interventions for patients with higher needs. 15,16,17,18,19,20,21,22 . Possible mechanisms for reduction in hospital-based healthcare utilization could be explained by the qualitative ndings of the larger study. Participants and caregivers reported that they were able to consolidate care at PCMH, as they received comprehensive care at PCMH that was of similar or better quality compared to their previous usual providers. The positive care experience was attributed to the continuity, personalisation and holistic management of care, and a sustained patient-provider relationship. These ndings were triangulated by ndings reported in other studies. A qualitative study with patients, payers, implementation staff and experts identi ed that continuity of care and adoption of care plan were amongst high-value elements that reduce healthcare utilization. Another cohort study observed that patients with continuity of care were less likely to have ED visits.
The one-year time lag before signi cant reductions was also consistent with the literature. A previous randomized controlled trial in the United States found that amongst patients at high risk for hospitalization, non-signi cant changes were observed in the rst year but signi cant reductions in inpatient admissions and ED visits were observed in the second year. The authors proposed that a period of engagement was needed to develop trust between the care team and the patients, before patients would reduce utilization of other healthcare services. 16 This study had various strengths. Firstly, this study used administrative data from a reliable public regional health system to calculate healthcare utilization. This afforded good coverage of the study participants' healthcare utilization, as public providers are major players in Singapore's healthcare scene, with public hospitals providing approximately 80% or more of tertiary inpatient services. 29 Administrative data also provided an accurate and objective measure that is not affected by recall error or other response biases.
Secondly, this analysis used a difference-in-difference approach. This design enabled assessment of whether enrolment into PCMH was associated with changes in healthcare utilization over time that were statistically different from the secular trend, combining the strengths of a case-control comparison and cohort study.
Thirdly, analysis of quarterly healthcare utilization provided more granular insights compared to commonly-seen measures of yearly healthcare utilization, while retaining su cient aggregation to tolerate 'noise' in the data.
There were also some limitations to this study. Firstly, the controls are matched only on available observables, and hence may not be ideal counterfactuals, as we did not have su cient data to match on psychological and social needs. Several indicators suggest that PCMH study participants had higher needs than controls. For example, PCMH study participants had signi cantly fewer polyclinic visits and more inpatient admissions and ED visits at baseline, compared to the controls. The higher inpatient admissions and ED visits at baseline were contributed by increases over 8 quarters or more before enrolment. This points to a prolonged increase in complexity of medical needs rather than an acute crisis episode, and is unlikely to be followed by recovery with usual care. . In addition, PCMH study participants had a higher prevalence of dementia (P:18%; C:6%), which is strongly associated with higher long-term psychosocial needs and healthcare utilization. , Accordingly, relative improvements after enrolment into PCMH would likely be larger if compared to a group with more similar needs.
Causal inference of the estimated reduction of polyclinic visits should also be interpreted with caution, since a small, marginally-signi cant reduction was observed even before enrolment into PCMH (Q −2 dy/dx: 0.24, p=0.076). This reduction occurred with concomitant increases in SOC visits, inpatient admissions and ED visits before enrolment. As ComSA-PCMH enrolled patients with complex care needs, it could be postulated that by the time of enrolment, polyclinic visits might have become inadequate to support the increase in complexity of care needs. Some care substitution with hospital services may have occurred.
In addition, this analysis did not differentiate between avoidable and unavoidable hospital visits. Avoidable visits include those due to social reasons or milder conditions that could be treated in primary care, or due to escalations that could have been avoided if the conditions had been better managed in primary care. As such, the reductions for avoidable visits could be expected to be more pronounced compared to reductions in all hospital visits, as PCMH is expected to reduce hospital visits via improved management of ambulatory-care sensitive conditions as well as consolidation and coordination of care. 15 This can be assessed in future research to improve understanding of the mechanisms that contributed to the reductions in health utilization.
Lastly, this analysis assessed healthcare utilization in a limited time period post-enrolment. Our ndings suggest that a longer follow-up period may allow us to capture more comprehensive effects.

Conclusions
Our study ndings support the hypothesis that enrolment into PCMH is associated with reductions in hospital-based healthcare utilization. While further research is needed to assess its longer-term effectiveness, sustainability and generalizability to other parts of Asia, these ndings add to a growing body of evidence supporting the bene ts of PCMHs for adults with complex needs globally and point to the potential of PCMH in Asia. NHG DSRB approved a waiver of consent for extraction of administrative data from the public regional health system, as study ndings would not change the care which the participants would have already received. Written informed consent was obtained from study participants who were not in the public regional health system's administrative records, for extraction of their records from the PCMH clinic administrative database.

Abbreviations
ii. Consent for publication Not applicable -this manuscript does not contain any personally-identi able information.
iii. Availability of data and materials The data presented in this study are not publicly available to protect patients' privacy.
iv. Competing interests CHW is currently employed at Tsao Foundation. The employment commenced after the completion of the study and analysis, and did not have any impact on manuscript preparation. All other authors had no con ict of interests relating to the subject matter discussed in this manuscript.

v. Funding
This study was funded by the Geriatric Education and Research Institute Intramural Research Grant (GERI1608). The funder played no part in study design, data collection, analysis, interpretation, or manuscript writing.