Healthcare utilization and cost trajectories post-stroke: role of caregiver and stroke factors

Singapore has a mixed healthcare delivery system where public, private and non-profit healthcare institutions deliver inpatient and ED services, intermediate and long-term care (ILTC) as well as specialist and primary ambulatory care. The public sector is the dominant provider of inpatient, ED and specialist outpatient services in tertiary hospitals accounting for more than 80% of market share in these services. The ILTC sector comprises of step-down care, nursing homes and hospice facilities. The sector also includes day services like day rehabilitation, home care services and others. Ministry of Health is the overarching regulatory body responsible for provision and regulation of healthcare services. Stroke patients present with acute symptoms at ED of any of the 5 public tertiary hospitals and are subsequently stabilized and transferred to ward. They undergo rehabilitation and following discharge, are referred to outpatient services to continue care in the community.

This was a prospective study in Singapore with recruitment period extending from December 2010 to September 2013. Stroke patients and their caregivers were recruited from all five tertiary hospitals of Singapore during that period, which ensured representativeness of our sample. Stroke patients were eligible if they were: (i) Singaporean or permanent resident, more than 40 years old and residing in Singapore for the next 1 year, (ii) stroke must be: recent diagnosis (i.e. stroke symptoms occurring within 4 weeks prior to admission) with diagnosis made by clinician and/or supported by brain imaging (CT or MRI) and (iii) not globally aphasic. A caregiver could be an immediate or extended family member or friend, more than 21 years (the legal definition of adult in Singapore), providing care or assistance of any kind and taking responsibility for the patient, as recognized by the patient and not fully paid for caregiving.

Within each of these tertiary hospitals, on-site research nurses reviewed the list of admitted stroke patients on a daily basis to screen eligible subjects and invited them to participate in the study. Written informed consent was obtained from all participants after explaining the study and procedures involved in the language they understood. They were informed that at any point during the follow-up period of 1 year, they can withdraw from the study, if they wished. Participants (including both the stroke patients and their caregivers) were interviewed via face-to-face interviews at baseline, 3-month and 12-month time points and only caregivers were interviewed via telephone interviews at 6-month and 9-month time points. Self-reported data was collected broadly under health, social and financial domains. To ensure good compliance, reminders were sent 1 week prior to the scheduled follow-up in the form of mails, phone calls or text messages. Interviews were scheduled over weekends or evenings of weekdays for participant’s convenience and adherence. Multiple attempts were made to contact participants before categorizing them as lost to follow-up. Training of first generation research assistants was conducted by main investigators, covering the content and appropriate method of data collection. The recordings of these training sessions were used to train subsequent research assistants to standardize data collection process. We pilot tested our survey with 40 participants from two of the five sites and made necessary amendments and finalized the survey forms.

For current study, while outcome variables were extracted from the national claims record, independent variables were taken from the survey conducted in Singapore Stroke Study at recruitment. National claims record is a nation-wide database of healthcare services utilization and associated costs maintained at the Ministry of Health, Singapore. All Singapore citizens and permanent residents are assigned a unique identification number which is used in almost all administrative contexts, including healthcare utilization. Linkage across multiple databases can be successfully accomplished using this identifier. For current study, adopting this approach the match rate was more than 95%.

Our main independent variable was quarter, a categorical variable representing each 3-month period post-stroke ranging from quarter 1 (Q1) to quarter 4 (Q4). Other covariates, collected from the study were socio-demographic, clinical, functional and caregiver related. The variables included under the socio-demographic group were: patient’s age, gender, ethnicity, religion, marital status, comorbid status and ward class (proxy for socio-economic status). Ward class referred to type of hospital ward stroke patient stayed in during the index stroke hospitalization. In Singapore, based on fulfilling means tested eligibility criteria, patients can be admitted in wards ranging from A, B1, B2 and C with different level of government subsidy. We binarized this ward class variable into two categories of subsidized and non-subsidized ward class.

Those under clinical group were: stroke type, recurrence and severity measured on the National Institute of Health Stroke Scale (NIHSS). Functional status was assessed using two scales, Modified Rankin Scale (mRS) and Barthel Index (BI) [38, 39]. No collinearity was observed between BI and mRS and both were included as covariates. Patients’ cognitive status was recorded using mini-mental state examination (MMSE) [40], while depression was assessed using Center for Epidemiological Studies Depression scale (CES-D-11 scale) [41]. Primary caregiver was an immediate or extended family member or friend, designated by the patient as the person providing physical care and their responses were categorized as none, spouse, child, sibling or others (includes distant relatives and friends).

HSU and associated cost data were extracted from the national claims record, over 3-monthly periods or quarters to study the trajectory and associated variation in HSU and associated costs over one-year post-stroke. Our HSU outcome variables were counts of visits per quarter for inpatient and ED service (excluding the index-stroke event), SOC and PC visits. We extracted total costs (inclusive of subsidies and out-of-pocket components) for inpatient, ED and PC visits.

Liang and Zeger [42] introduced the Generalized Estimating Equation (GEE) approach for analyzing data with repeated measures for Generalized Linear Models which provides population average estimates. For the count data, we chose poisson or negative binomial distribution with a log link function pending on whether there was evidence of over-dispersion [43]. Gamma distribution with a log link was used for cost data. We specified working correlation matrix corresponding to an autoregressive order 1 correlation structure and used the Huber and White sandwich estimator to obtain robust variance estimates even if the working correlation matrix is misspecified [42, 44, 45].

Compared to first quarter post-stroke, the unadjusted incidence rate ratio (IRR) of acute hospitalization in subsequent three quarters was 0.74 (95% CI: 0.54, 1.02), 0.65 (95% CI: 0.48, 0.88) and 0.67 (95% CI: 0.47, 0.96) respectively. The IRR estimates remained almost unchanged after adjusting for covariates. A decreasing trend of acute hospitalization associated cost was observed with adjusted exponentiated cost estimates from Q2 to Q4 being 0.52 (95% CI: 0.32, 0.83), 0.40 (95% CI: 0.23, 0.68) and 0.33 (95% CI: 0.20, 0.54) when compared with Q1 suggesting expected acute hospitalization associated cost decreased by 48, 60 and 67% for Q2, Q3 and Q4 respectively when contrasted with Q1. Both adjusted acute hospitalization and associated cost post-stroke showed a statistically significant linear trend effect that decreased with time since stroke, with greater decrease in the cost compared to number of visits (Fig. 1). Compared to first quarter, the adjusted IRR of ED service use in subsequent quarters was 0.91 (95% CI: 0.70, 1.19), 0.76 (95% CI: 0.58, 0.99) and 0.82 (95% CI: 0.57, 1.17) respectively. Both ED visit and associated cost showed a decreasing trend till third quarter post-stroke, but the linear trend effect was not statistically significant (Fig. 1). Compared to first quarter post-stroke, the unadjusted IRR of SOC service use in subsequent three quarters were 1.00 (95% CI: 0.89, 1.13), 0.75 (95% CI: 0.67, 0.84) and 0.74 (95% CI: 0.65, 0.85) respectively. After adjustment, the IRR of SOC service use had a similar significant linear trend with utilization still being the highest in the first 6 months and decreasing subsequently (Fig. 1). PC service utilization had a significant overall trend where the highest utilization was in the quarter immediately post-stroke and then decreased subsequently with adjusted IRR of PC visits per quarter being 0.93 (95% CI: 0.81, 1.06), 0.93 (95% CI: 0.79, 1.09) and 0.78 (95% CI: 0.67, 0.90) respectively in Q2 to Q4 when compared with Q1. The PC service associated cost showed a significant linear trend that had an opposite trend to PC visits with adjusted, exponentiated cost estimates in Q2, Q3 and Q4 being 1.33 (95% CI: 1.14, 1.54), 1.35 (95% CI: 1.15, 1.59) and 1.36 (95% CI: 1.15, 1.62) when compared with Q1 suggesting expected PC associated cost increased by 33, 35 and 36% for Q2, Q3 and Q4 respectively when contrasted with Q1 (Fig. 1). While both cost and service usage decreased across post-stroke quarters for inpatient and ED services, the service usage decreased, and cost increased across post-stroke quarters for PC services.

Caregiver identity was associated with acute inpatient and ED visits and their associated costs only. The IRR of acute hospitalization decreased by 51, 40, 11 and 1% for patients having spousal, sibling, child and others as caregivers respectively when compared with those having no caregiver (p = 0.017). Like hospitalization, acute hospitalization related costs for those with any related caregiver were lower than those with no caregiver. Having a spousal caregiver was associated with the lowest acute hospitalization costs with exponentiated cost estimate being 0.16 (95% CI: 0.09, 0.29) when compared with no caregivers suggesting expected hospitalization cost decreased by 84% when contrasted with no caregivers (p < 0.001) (Fig. 2).

The interaction term between quarter and disability (measured on mRS) was statistically significant (p < 0.001) in SOC service utilization model suggesting the utilization trend was different among varying levels of disability in stroke patients. Though the SOC utilization trend was decreasing across post-stroke quarters for both no or slightly disabled (mRS: 0–2) and moderate or severely disabled groups (mRS: 3–5), the decrease was steeper for the group with no or slight disability (Fig. 3) where the linear trend of SOC utilization across 4 quarters post-stroke in both groups were statistically significant (Table 3).

The PC costs across post-stroke quarters varied by stroke type and severity. Figure 4(a) illustrates the PC cost trend having an increasingly pronounced positive linear trend as we progressed from mild, moderately severe to severe ischemic stroke subgroups although the linear trend of PC associated cost was statistically significant only for moderately severe ischemic group with borderline significance (p < 0.15) for severe ischemic group (Table 3). Figure 4(b) illustrates the PC cost trend for mild, moderately severe and severe non-ischemic stroke subgroups. For all three groups, compared to cost in Q1, there was an increase in Q2, Q3 after which the increase in Q4 was slightly lower than Q3. However, the magnitude of increase across Q2 to Q4, compared to Q1 was increasingly pronounced as stroke severity increased where the linear trend for PC service associated costs was statistically significant for moderately severe and severe non-ischemic groups with borderline significance (p < 0.15) for mild non-ischemic group (Table 3). The overall results are summarized in Table 4.