Abstract
Background
High-cost users (HCU) represent important targets for health policy interventions. Sepsis is a life-threatening syndrome that is associated with high morbidity, mortality, and economic costs to the healthcare system. We sought to estimate the effect of sepsis on being a subsequent HCU.
Methods
Using linked health-administrative databases, we conducted a population-based, propensity score-weighted cohort study of adults who survived a hospitalization in Ontario, Canada between January 2016 and December 2017. Sepsis was identified using a validated algorithm. The primary outcome was being a persistent HCU after hospital discharge (in the top 5% or 1% of total health care spending for 90 consecutive days), and the proportion of follow-up time since discharge as a HCU.
Results
We identified 927,057 hospitalized individuals, of whom 79,065 had sepsis. Individuals who had sepsis were more likely to be a top 5% HCU for 90 consecutive days at any time after discharge compared to those without sepsis (OR 2.24; 95% confidence interval [CI] 2.04–2.46) and spent on average 42.3% of their follow up time as a top 5% HCU compared to 28.9% of time among those without sepsis (RR 1.46; 95% CI 1.45–1.48). Individuals with sepsis were more likely to be a top 1% HCU for 90 consecutive days compared to those without sepsis (10% versus 5.1%, OR 2.05 [95% CI 1.99–2.11]), and spent more time as a top 1% HCU (18.5% of time versus 10.8% of time, RR 1.68 [95% CI 1.65–1.70]).
Conclusions
The sequelae of sepsis result in higher healthcare costs with important economic implications. After discharge, individuals who experienced sepsis are more likely to be a HCU and spend more time as a HCU compared to individuals who did not experience sepsis during hospitalization.
Supplementary Information
The online version contains supplementary material available at 10.1186/s13054-024-05108-6.
Keywords: Sepsis, High-cost users, Healthcare utilization, Healthcare costs
Background
Sepsis is a global health threat responsible for an estimated 20% of deaths worldwide1. This life-threatening health condition is also associated with economic costs in high-, middle-, and low-income countries and is associated with significant morbidity [1–6]. Sepsis survivors experience an increased risk of cardiovascular disease, long-term cognitive and functional impairment, worsening of pre-existing health chronic health conditions, and an increased risk of rehospitalization and death [2, 4, 6–8]. Sepsis and its long-term consequences are associated with significant costs to healthcare systems [9–11].
Individuals who have the highest health care needs in a population, often referred to as “High-Cost High-Need” or “High-Cost Users” (HCUs) represent a small percentage of the overall population, but account for a disproportionate amount of healthcare spending. Canadian [12], US [13], British [14], and Australian [15] studies have all shown that the top 5% of the population account for about 50% of total healthcare dollars spent [16]. These costs are highly skewed: a Canadian study reported that the top 1% of healthcare users accounted for 27.5% of total healthcare spending, as compared to the bottom 50% who incurred less than 4% of total healthcare costs [12]. Interventions aimed at preventing individuals from becoming a HCU by improving health and well-being could result in significant cost savings to the health system. Previous research has sought to understand the interplay between social determinants of health and becoming a HCU [17–19], but less is known about how specific health events create incident and persistent HCUs.
Although the long-term health consequences and costs after sepsis are well described [4, 6, 7], it is not known if individuals who survive a hospitalization with sepsis are more likely to become a HCU after discharge. The objective of this study was to determine if an episode of sepsis during a hospitalization increases an individual’s risk of becoming a HCU, compared to a hospitalization without sepsis.
Methods
Setting, population
The setting for this study is the province of Ontario (population 16 million) [12], which has a publicly funded, universal healthcare system. This population-based, propensity score-weighted cohort study was conducted using individual-level health administrative datasets for the province of Ontario, which contain records for all acute care hospitalizations in the province [13]. Individual records are linked using unique encoded identifiers and analyzed at ICES. Details regarding dataset validation and data access protocols are reported elsewhere [13]. ICES is an independent, non-profit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyze health care and demographic data, without consent, for health system evaluation and improvement. The use of the data in this project is authorized under Sect. 45 of Ontario’s Personal Health Information Protection Act (PHIPA) and does not require review by a Research Ethics Board. (Please see Supplementary Material 1: Appendix 1 for a description of ICES datasets used). The results of our study are reported according to the Reporting of studies Conducted using Observational Data (RECORD). [14]
Cohort creation
We identified all adults aged ≥ 18 years who were eligible for publicly funded healthcare, who had no previous diagnosis of sepsis during a 2-year lookback period, and who survived an acute, non-obstetric hospitalization between January 1, 2016, and December 31, 2017. For individuals who had multiple hospitalizations during this accrual period, we selected the first. Hospitalization admission date was used as the index date, from which baseline covariates were collected and the look-back started. We considered the hospitalization to be the exposure, and we therefore did not include time during the hospitalization in the observed follow-up period. Follow-up time started the day after discharge from the hospital, and individuals were followed until death or administrative censoring on December 31, 2019, whichever was earliest.
Exposure
Among our hospitalized cohort, we identified individuals exposed to sepsis using a previously described algorithm that uses ICD-10 codes for sepsis, organ dysfunction, and infection in the discharge abstract database [15]. (Please see Supplementary Material 1: Appendix 2 for details). Individuals who did not have sepsis identified during their index hospitalization were the unexposed comparator population.
Covariates
We identified the following covariates at index date: age, sex, rurality, neighbourhood income quintile, area-level ecological markers of social determinants of health using the Ontario Marginalization index [16], and if the individual had immigrated to Canada prior to index date. We used a one-year lookback to identify markers of previous healthcare resource utilization including previous hospitalizations, emergency department visits, physician visits, healthcare costs, and whether individuals’ healthcare costs were in the top 10% of spending for each 30-day period, (see methods for healthcare costs below). We used a two-year lookback to identify whether the individual had resided in a long-term care home, and to identify the presence of medical comorbidities that are known to predispose to sepsis and healthcare resource utilization (cancer, congestive heart failure, chronic obstructive pulmonary disease, chronic kidney disease or chronic dialysis, and diabetes) [17]. We used the Johns Hopkins ACG® System Version 10, which is a multi-morbidity framework that groups individuals based on expected future healthcare resource utilization, to identify frailty, resource utilization bands, and aggregated diagnosis groups (ADGs). [18–20]
We identified the following covariates for the index hospitalization: hospital size, type of hospitalization based on case mix groups (medical, surgical, other), Charlson score, hospital length of stay, admission to an intensive care unit (ICU), ICU length of stay, receipt of mechanical ventilation (non-invasive and invasive), receipt of extracorporeal life support (ECLS), healthcare costs for the index hospitalization, and discharge location. See Supplementary Material 1: Appendix 3 for a list of sources for all covariates.
Propensity score weighting
We calculated the propensity score by regressing sepsis exposure on 17 baseline covariates selected based on previous research and our clinical understanding of the risk factors for sepsis and for future healthcare resource utilization (see Supplementary Material 1: Appendix 4) [2, 21–25]. Hospital and ICU length of stay, and the need for mechanical ventilation were not included as covariates in the propensity score model, as we viewed these as mediators rather than confounding factors. Adding mediators to IPTW alongside confounders can lead to overadjustment bias, induce collider bias, and distort the effect estimation. We did not believe ICU admission was a mediator and therefore we did include this covariate in the propensity score model. To adjust for confounding, we calculated inverse probability of treatment weights (IPTW) for each individual using their propensity score and exposure status. [26, 27] Compared to propensity score matching, the use of IPTW to control for confounding has several advantages including inclusion of all individuals in the dataset, ability to achieve balance across a range of covariates, and improved overall precision [26]. We assessed the balance of covariates between the two groups before and after weighting by comparing the weighted standardized difference (WSD), using an acceptable threshold of < 10%.
Outcomes
We estimated total healthcare costs for each individual in our cohort in 30-day intervals from hospital discharge to the end of follow-up using a hybrid costing methodology developed for administrative databases. [28, 29] The algorithm identifies healthcare encounters such as physician visits, diagnostic tests, same-day surgery, hospitalizations, emergency department visits, rehabilitation, homecare, prescription drugs, complex continuing care, and long-term care. Costs were calculated for each encounter using the unit cost or by multiplying the resource intensity weights specific to each database by the average provincial cost per weighted case. Costs were standardized to 2020 Canadian dollar values. [29]
The primary outcome was HCU status after hospitalization. HCU status was identified by comparing individual healthcare costs per 30-day period to the pre-identified top 5% cost threshold specific to that same 30-day calendar period (further details available in Supplementary Material 1: Appendix 5 and 6). Individuals whose 30-day costs were at or above the thresholds were a HCU compared to total population health spending for that time-period. We report the primary outcome as being a persistent HCU (the top 5%) for any period of 90 consecutive days (3 consecutive 30-day periods) after discharge from hospital. We also report the proportion of individual subjects’ follow-up time spent as a HCU since hospital discharge, which was calculated using individual duration of follow-up time as the denominator. We repeated all analyses defining HCU at the top 1% threshold, which is a more specific definition of HCUs, and places individuals among the very top of healthcare spending.
Subgroup analyses
Previous studies exploring trajectories of HCU in Ontario have demonstrated that almost half of individuals in the top 10% of healthcare spending continue to be in the top 10% in subsequent years. [25] We therefore expected that many individuals in our cohorts would be pre-existing HCUs. To better understand the impact of sepsis among those who were not previously a HCU, we conducted additional analysis stratifying individuals based on whether they had ever been in the top 10% of healthcare spending for 30-days in the year prior to hospitalization. We recalculated the propensity score for sepsis within each strata (previous top 10%, not previously in top 10%) and repeated IPTW weighting within each strata. Baseline characteristics of these two stratified cohorts before and after weighting are presented in Supplementary Material 1: Appendix 7.
Statistical analyses
We used a weighted logistic regression with robust standard error to estimate the odds ratio (OR) of becoming a persistent HCU (any period of 90 consecutive days as a HCU) comparing individuals who were hospitalized with sepsis to those who did not have sepsis. We used weighted Poisson regression modeling with robust standard error to estimate the relative rate of being a HCU at any point during follow-up time. All statistical analyses were done using SAS® software, Version 9.4 (SAS Institute Inc., Cary, NC, USA).
Results
There were 927,057 adults who survived an admission to an Ontario hospital for a non-obstetric reason in 2016 and 2017, among whom we identified an episode of sepsis in 79,065 (see Table 1 and Fig. 1). Individuals experiencing sepsis were older (mean age 72.8 [SD 16.3] versus 60.2 [SD 19.1]) years old), more likely to be frail (22.3% versus 8.2%), and more likely to have had a hospitalization in the year prior (8.8% versus 6.3%). Almost all (95.0% sepsis versus 92.3% non-sepsis) individuals in the cohort had at least one 30-day period during the 2 years prior to the hospitalization in which their 30-day total healthcare cost was in the top 10%. After weighting, baseline characteristics were balanced between the two cohorts for all covariates, except for frailty (12.9% sepsis versus 9.24% non-sepsis [WSD 0.12]), and among those in the highest band of previous resource utilization (27.9% sepsis versus 23.3% non-sepsis [WSD 0.11]).
Table 1.
Baseline characteristics of individuals with and without sepsis during hospitalization
| Variable | Prior to weighting | After weighting | ||||
|---|---|---|---|---|---|---|
| Sepsis N = 79, 065 | Non-Sepsis N = 847, 992 | Standardized Difference | Sepsis N = 888,513 | Non-Sepsis N = 928,323 | Weighted Standardized Difference | |
| Age, mean ± SD median (IQR) |
72.77 ± 16.25 76 (64–85) |
60.19 ± 19.06 63 (47–75) |
0.71 0.72 |
62.14 | 61.29 | 0.044 |
| Female, n (%) | 44,556 (56.4%) | 457,932 (54.0%) | 0.05 | 52.01% | 54.19% | 0.044 |
| Immigration Status, n (%) | ||||||
| Immigrated prior to index date | 6,496 (8.2%) | 104,705 (12.3%) | 0.14 | 11.13% | 11.98% | 0.026 |
| Non-immigrant | 72,569 (91.8%) | 743,287 (87.7%) | 0.14 | 88.87% | 88.02% | 0.026 |
| Recent immigrant (≤ 5 years) | 387 (0.5%) | 10,701 (1.3%) | 0.08 | 0.83% | 1.20% | 0.037 |
| Long-term immigrant (> 5 years) | 6,109 (7.7%) | 94,004 (11.1%) | 0.12 | 10.30% | 10.78% | 0.016 |
| Rural, n (%) | 7,834 (9.9%) | 82,088 (9.7%) | 0.01 | 9.87% | 9.70% | 0.006 |
| Neighborhood Income Quintile, n (%) | ||||||
| 1 (lowest) | 20,798 (26.3%) | 187,527 (22.1%) | 0.1 | 24.10% | 22.50% | 0.038 |
| 2 | 17,439 (22.1%) | 177,345 (20.9%) | 0.03 | 21.49% | 21.00% | 0.012 |
| 3 | 14,993 (19.0%) | 167,404 (19.7%) | 0.02 | 19.45% | 19.66% | 0.005 |
| 4 | 13,262 (16.8%) | 157,678 (18.6%) | 0.05 | 17.92% | 18.42% | 0.013 |
| 5 (highest) | 12,200 (15.4%) | 155,285 (18.3%) | 0.08 | 16.66% | 18.09% | 0.038 |
| Missing information | 373 (0.5%) | 2,753 (0.3%) | 0.02 | 0.37% | 0.35% | 0.004 |
| Ontario marginalization index dimensions | ||||||
| Household and Dwellings, n (%) | ||||||
| 1 (least deprived) | 10,170 (12.9%) | 143,863 (17.0%) | 0.12 | 15.39% | 16.62% | 0.034 |
| 2 | 12,753 (16.1%) | 153,245 (18.1%) | 0.05 | 17.34% | 17.89% | 0.014 |
| 3 | 14,565 (18.4%) | 161,287 (19.0%) | 0.02 | 18.44% | 18.96% | 0.013 |
| 4 | 16,912 (21.4%) | 169,093 (19.9%) | 0.04 | 20.35% | 20.07% | 0.007 |
| 5 (most deprived) | 23,434 (29.6%) | 210,407 (24.8%) | 0.11 | 27.10% | 25.22% | 0.043 |
| Missing information | 1,231 (1.6%) | 10,097 (1.2%) | 0.03 | 1.38% | 1.23% | 0.013 |
| Material Resources, n (%) | ||||||
| 1 (least deprived) | 13,434 (17.0%) | 166,239 (19.6%) | 0.07 | 18.33% | 19.37% | 0.026 |
| 2 | 14,457 (18.3%) | 165,606 (19.5%) | 0.03 | 18.77% | 19.40% | 0.016 |
| 3 | 14,829 (18.8%) | 162,359 (19.1%) | 0.01 | 19.08% | 19.12% | 0.001 |
| 4 | 16,126 (20.4%) | 166,961 (19.7%) | 0.02 | 20.14% | 19.75% | 0.010 |
| 5 (most deprived) | 18,988 (24.0%) | 176,730 (20.8%) | 0.08 | 22.31% | 21.13% | 0.029 |
| Missing information | 1,231 (1.6%) | 10,097 (1.2%) | 0.03 | 1.38% | 1.23% | 0.013 |
| Age and Labour Force, n (%) | ||||||
| 1 (least deprived) | 10,888 (13.8%) | 162,257 (19.1%) | 0.15 | 17.68% | 18.68% | 0.026 |
| 2 | 12,413 (15.7%) | 153,407 (18.1%) | 0.06 | 17.70% | 17.90% | 0.005 |
| 3 | 13,190 (16.7%) | 149,258 (17.6%) | 0.02 | 17.62% | 17.51% | 0.003 |
| 4 | 15,324 (19.4%) | 157,636 (18.6%) | 0.02 | 18.83% | 18.63% | 0.005 |
| 5 (most deprived) | 26,019 (32.9%) | 215,337 (25.4%) | 0.17 | 26.80% | 26.06% | 0.017 |
| Missing information | 1,231 (1.6%) | 10,097 (1.2%) | 0.03 | 1.38% | 1.23% | 0.013 |
| Racialized and Newcomer Populations, n (%) | ||||||
| 1 (least deprived) | 18,166 (23.0%) | 177,230 (20.9%) | 0.05 | 21.28% | 21.07% | 0.005 |
| 2 | 16,646 (21.1%) | 164,547 (19.4%) | 0.04 | 20.20% | 19.49% | 0.018 |
| 3 | 14,442 (18.3%) | 155,651 (18.4%) | 0 | 18.10% | 18.36% | 0.007 |
| 4 | 13,544 (17.1%) | 159,303 (18.8%) | 0.04 | 17.79% | 18.69% | 0.023 |
| 5 (most deprived) | 15,036 (19.0%) | 181,164 (21.4%) | 0.06 | 21.26% | 21.16% | 0.002 |
| Missing information | 1,231 (1.6%) | 10,097 (1.2%) | 0.03 | 1.38% | 1.23% | 0.013 |
| Resource Utilization Bands, n (%) | ||||||
| Band 0 (non-users) | 926 (1.2%) | 9,448 (1.1%) | 0.01 | 1.44% | 1.09% | 0.031 |
| Band 1 | 457 (0.6%) | 5,536 (0.7%) | 0.01 | 0.92% | 0.63% | 0.033 |
| Band 2 | 2,314 (2.9%) | 34,099 (4.0%) | 0.06 | 4.14% | 3.90% | 0.012 |
| Band 3 | 25,731 (32.5%) | 327,205 (38.6%) | 0.13 | 36.94% | 38.05% | 0.023 |
| Band 4 | 22,316 (28.2%) | 283,066 (33.4%) | 0.11 | 28.64% | 33.09% | 0.096 |
| Band 5 (very high users) | 27,321 (34.6%) | 188,638 (22.2%) | 0.28 | 27.92% | 23.25% | 0.107 |
| Frailty indicator (ACGs), n (%) | 17,671 (22.3%) | 69,630 (8.2%) | 0.4 | 12.92% | 9.24% | 0.118 |
|
Number of ACGs, mean ± SD, median (IQR) |
8.67 ± 4.04 9 (6–11) |
8.00 ± 3.79 8 (5–11) |
0.17 0.17 |
8.33 | 8.06 | 0.067 |
| Number of ACGs (0), n (%) | 928 (1.2%) | 9,470 (1.1%) | 0.01 | 1.44% | 1.09% | 0.031 |
| Number of ACGs (1–9), n (%) | 45,302 (57.3%) | 554,494 (65.4%) | 0.17 | 60.56% | 64.72% | 0.086 |
| Number of ACGs (10–19), n (%) | 32,574 (41.2%) | 282,299 (33.3%) | 0.16 | 37.73% | 33.94% | 0.079 |
| Number of ACGs (20–29), n (%) | 261 (0.3%) | 1,729 (0.2%) | 0.02 | 0.27% | 0.26% | 0.003 |
| Prior Cancer, n (%), | 18,260 (23.1%) | 153,062 (18.0%) | 0.13 | 19.40% | 18.49% | 0.023 |
| Prior Congestive Heart Failure, n (%) | 14,711 (18.6%) | 74,383 (8.8%) | 0.29 | 10.78% | 9.67% | 0.037 |
| Prior chronic kidney disease or chronic dialysis, n (%) | 8,571 (10.8%) | 47,604 (5.6%) | 0.19 | 7.22% | 6.09% | 0.045 |
| Prior chronic obstructive pulmonary disease, n (%) | 26,963 (34.1%) | 159,031 (18.8%) | 0.35 | 22.07% | 20.12% | 0.048 |
| Prior diabetes, n (%) | 28,374 (35.9%) | 203,236 (24.0%) | 0.26 | 26.82% | 25.04% | 0.041 |
| Healthcare Use 1-year Prior Index | ||||||
| ED visits, mean ± SD, median (IQR) |
1.57 ± 2.51 1 (0–2) |
1.26 ± 2.60 1 (0–2) |
0.12 0.25 |
1.54 | 1.41 | 0.027 |
| Receipt of home care, n (%) | 27,260 (34.5%) | 132,322 (15.6%) | 0.45 | 19.75% | 17.32% | 0.063 |
| Hospitalization, n (%) | 6,983 (8.8%) | 53,320 (6.3%) | 0.1 | 7.51% | 6.55% | 0.038 |
| No. of hospitalization episodes, mean ± SD, median (IQR) |
0.13 ± 0.49 0 (0–0) |
0.09 ± 0.44 0 (0–0) |
0.08 0.1 |
0.11 | 0.1 | 0.026 |
| No. of physician visits, mean ± SD, median (IQR) |
11.52 ± 10.61 9 (5–15) |
11.36 ± 9.54 9 (5–15) |
0.02 0.02 |
12.14 | 11.39 | 0.064 |
| LTC resident in 2-year lookback, n (%) | 6,443 (8.1%) | 15,227 (1.8%) | 0.3 | 2.86% | 2.40% | 0.029 |
| Previous top 10%, n (%) | 75,105 (95.0%) | 782,399 (92.3%) | 0.11 | 93.14% | 92.51% | 0.024 |
SD Standard Deviation, IQR Interquartile Range, ACG Adjusted Clinical Groups, Frailty, and Resource utilization bands from the Johns Hopkins ACG® System Version 10, OnMARG Ontario Marginalization Index, LTC Long-term Care Resident
Fig. 1.
Cohort Creation Flow Diagram
About half (51.3%) of hospitalizations with sepsis were to large hospitals with over 100 beds (Table 2). Many sepsis hospitalizations (60.6%) were classified as medical, while the non-sepsis hospitalizations were equally distributed between medical, surgical, and other categories. Even after propensity score weighting, individuals with sepsis had longer mean hospital length of stay (12.8 versus 5.05 days [WSD 0.381]), were more frequently admitted to the ICU (17.6% versus 15.0% [WSD 0.069]), had longer mean ICU length of stay (9.95 versus 3.15 days [WSD 0.510]) and were more likely to receive mechanical ventilation (9.41% versus 4.26% [WSD 0.205]). The average cost of the index hospitalization in the weighted cohort was also higher among those with sepsis ($17,395 versus $7,925 [WSD 0.343]).
Table 2.
Characteristics of index hospitalization among individuals with and without sepsis
| Variable | Prior to weighting | After weighting | ||||
|---|---|---|---|---|---|---|
| Sepsis N = 79, 065 | Non-Sepsis N = 847, 992 | Standardized Difference | Sepsis N = 888,513 | Non-Sepsis N = 928,323 | Weighted Standardized Difference | |
| Hospital Size, n (%) | ||||||
| Teaching | 20,511 (25.9%) | 279,789 (33.0%) | 0.16 | 27.41% | 32.85% | 0.119 |
| Community ≥ 100 beds | 40,551 (51.3%) | 424,804 (50.1%) | 0.02 | 51.30% | 50.06% | 0.025 |
| Community < 100 beds | 16,758 (21.2%) | 128,145 (15.1%) | 0.16 | 19.77% | 15.28% | 0.118 |
| Missing information | 1,245 (1.6%) | 15,254 (1.8%) | 0.02 | 1.51% | 1.81% | 0.023 |
| Type of hospitalization, n (%) | ||||||
|
Medical (including mental health) |
47,903 (60.6%) | 287,902 (34.0%) | 0.55 | 60.51% | 34.61% | 0.537 |
| Surgical | 8,159 (10.3%) | 291,135 (34.3%) | 0.6 | 10.76% | 33.89% | 0.578 |
| Not categorized | 23,003 (29.1%) | 268,955 (31.7%) | 0.06 | 28.74% | 31.51% | 0.060 |
| Charlson score of index hospitalization, mean ± SD, median (IQR) |
1.51 ± 1.68 1 (0–2) |
0.89 ± 1.46 0 (0–1) |
0.39 0.52 |
1.24 | 0.92 | 0.207 |
| Length of Stay (days), mean ± SD, median (IQR) |
14.79 ± 28.89 7 (4–15) |
4.88 ± 9.91 3 (1–5) |
0.46 0.96 |
12.82 | 5.05 | 0.381 |
| ICU Admission during index admission, n (%) | 16,750 (21.2%) | 121,991 (14.4%) | 0.18 | 17.57% | 15.03% | 0.069 |
| ICU length of stay (days), mean ± SD, median (IQR) |
8.84 ± 17.42 5 (2–9) |
3.14 ± 4.96 2 (1–4) |
0.45 0.81 |
9.95 | 3.15 | 0.510 |
| Mechanical ventilation, n (%) | 8,301 (10.5%) | 35,006 (4.1%) | 0.25 | 9.41% | 4.26% | 0.205 |
| ECLS during index admission, n (%) | 151 (0.2%) | 354 (0.0%) | 0.04 | 0.21% | 0.04% | 0.046 |
| Discharge location, n (%) | ||||||
| Transferred to acute inpatient facility | 955 (1.2%) | 6,730 (0.8%) | 0.04 | 1.19% | 0.81% | 0.038 |
|
Transferred to long-term care, continuing care, or supportive housing |
17,867 (22.6%) | 64,046 (7.6%) | 0.43 | 14.51% | 8.37% | 0.194 |
| Discharged home | 59,559 (75.3%) | 767,584 (90.5%) | 0.41 | 82.80% | 89.72% | 0.202 |
| Discharged other [e.g., AMA, on pass] | 684 (0.9%) | 9,632 (1.1%) | 0.03 | 1.51% | 1.11% | 0.035 |
| Cost of index hospitalization, Mean ± SD, Median (IQR) |
$18,434.82 ± $37,142.81 $8,713 ($5,486—$16,687) |
$7,788.10 ± $10,962.88 $5,212 ($3,269—$8,372) |
0.39 0.78 |
$17,394.6 | $7924.88 | 0.343 |
SD Standard Deviation, IQR Interquartile Range, ECLS Extracorporeal life support
aThis includes invasive or non-invasive mechanical ventilation
Compared to individuals who did not have sepsis during their index hospitalization, those who had sepsis were more likely to have healthcare spending that places them in the top 5% (OR 1.73 [95% CI 1.70–1.77]) or top 1% (OR 2.05 [95% CI 1.99–2.11]) of healthcare spending for 90 consecutive days at any time during follow up (Table 3).
Table 3.
High-cost user status post-discharge for those with and without sepsis during hospitalization
| Variable | Unweighted | Weighted | ||||
|---|---|---|---|---|---|---|
| Sepsis | Non-Sepsis | Measure of Association | Sepsis | Non-Sepsis | Measure of Association | |
| Total cohort | ||||||
| N = 79, 065 | N = 847, 992 | N = 888,513 | N = 928,323 | |||
| Death during follow-up, n (%) | 30,158 (38.1%) | 127,925 (15.1%) | 25.60% | 16.41% | ||
| High-Cost User (Top 5%) for any 90-day period after discharge from hospital, n (%) | 35,419 (44.8%) | 201,533 (23.8%) | OR 2.60 (95% CI 2.56, 2.64) | 36.51% | 24.91% | OR 1.73 (95% CI 1.70, 1.77) |
| High-Cost User (Top 1%) for any 90-day period after discharge from hospital, n (%) | 10,082 (12.8%) | 40,282 (4.8%) | OR 2.93 (95% CI 2.86, 3.00) | 9.96% | 5.13% | OR 2.05 (95% CI 1.99, 2.11) |
| Proportion of follow-up time as a High-Cost User (Top 5%), Mean ± SD, Median (IQR) |
55.04% ± 39.22 52% (15–100) |
27.33% ± 33.18 12% (3–39) |
RR 2.01 (95% CI 2.01, 2.02) | 42.27% | 28.89% | RR 1.46 (95% CI 1.45, 1.48) |
| Proportion of follow-up time as a High-Cost User (Top 1%), Mean ± SD, Median (IQR) |
24.48 ± 33.11 8 (1–33) |
10.09 ± 22.33 2 (0–7) |
RR 2.43 (95% CI 2.42, 2.43) | 18.15% | 10.83% | RR 1.68 (95% CI 1.65, 1.70) |
| Total length of follow-up (days), Mean ± SD, Median (IQR) |
839.51 ± 437.37 934 (531 – 1,189) |
1,016.06 ± 342.51 1,063 (854 – 1,281) |
930.17 | 1006.42 | ||
| Previously in top 10% | ||||||
| N = 75, 105 | N = 782, 399 | N = 821,753 | N = 858,653 | |||
| High-Cost User (Top 5%) for any 90-day period after discharge from hospital, n (%) | 34,275 (45.6%) | 194,921 (24.9%) | OR 2.53 (95% CI 2.49, 2.57) | 37.62% | 26.07% | OR 1.71 (95% CI 1.68, 1.74) |
| High-Cost User (Top 1%) for any 90-day period after discharge from hospital, n (%) | 9,673 (12.9%) | 38,716 (4.9%) | OR 2.84 (95% CI 2.77, 2.91) | 10.14% | 5.33% | OR 2.00 (95% CI 1.94, 2.07) |
| Proportion of follow-up time as a High-Cost User (Top 5%), Mean ± SD, Median (IQR) |
56.40% ± 38.88 56% (17–100) |
28.78% ± 33.62 13% (4–43) |
RR 1.96 (95% CI 1.96, 1.96) | 43.94% | 30.37% | RR 1.45 (95% CI 1.44, 1.46) |
| Proportion of follow-up time as a High-Cost User (Top 1%), Mean ± SD, Median (IQR) |
25.04 ± 33.31 9 (2–35) |
10.62 ± 22.84 2 (0–8) |
RR 2.36 (95% CI 2.35, 2.36) | 18.8% | 11.38% | RR 1.65 (95% CI 1.63, 1.68) |
| Not previously in top 10% | ||||||
| N = 3, 960 | N = 65, 593 | N = 66,961 | N = 69,578 | |||
| High-Cost User (Top 5%) for any 90-day period after discharge from hospital, n (%) l | 1,144 (28.9%) | 6,612 (10.1%) | OR 3.62 (95% CI 3.37, 3.90) | 20.89% | 10.54% | OR 2.24 (95% CI 2.04, 2.46) |
| High-Cost User (Top 1%) for any 90-day period after discharge from hospital, n (%) | 409 (10.3%) | 1,566 (2.4%) | OR 4.71 (95% CI 4.20, 5.28) | 7.13% | 2.50% | OR 2.99 (95% CI 2.60, 3.45) |
| Proportion of follow-up time as a High-Cost User (Top 5%), Mean ± SD, Median (IQR) |
29.31% ± 36.79 10% (3–46) |
10.05% ± 20.71 3% (0–8) |
RR 2.92 (95% CI 2.90, 2.94) | 19.39% | 10.55% | RR 1.84 (95% CI 1.74, 1.94) |
| Proportion of follow-up time as a High-Cost User (Top 1%), Mean ± SD, Median (IQR) |
13.93 ± 26.96 3 (0–11) |
3.79 ± 13.48 0 (0–2) |
RR 3.68 (95% CI 3.64, 3.71) | 9.13% | 4.01% | RR 2.28 (95% CI 2.10, 2.48) |
OR Odds Ratio, RR Relative Rate, SD Standard Deviation, IQR Interquartile Range
Individuals who had sepsis during their hospitalization spent on average 42.3% of their follow up time as a top 5% HCU, compared to 28.9% among those without sepsis (RR 1.46 [95% CI 1.45–1.48]). They spent on average 18.2% of their follow-up time as a top 1% of HCU, compared to 10.8% among those without sepsis (RR 1.68 [95% CI 1.65–1.70]).
In the stratified analysis including only those who were not previously in the top 10% of health care costs, individuals with sepsis were a top 5% HCU for 19.4% of their observation time compared to 10.6% among those who did not have sepsis (RR 1.84 [95% CI 1.74–1.94]). The odds of being a top 1% HCU for 90 days after discharge was highest among this group (OR 2.99 [95% CI 2.60, 3.45]).
Discussion
This study demonstrates that individuals who experience sepsis during a hospitalization are more likely to be a top 5% or top 1% high-cost user for any consecutive 90-day period after discharge and spend more time after discharge as a HCU compared to individuals hospitalized without sepsis. The odds of becoming a HCU after sepsis are highest among those who did not previously have high health-care costs, suggesting that sepsis can significantly change the trajectory of an individual’s healthcare utilization. We also demonstrate that hospitalizations with sepsis are more resource intensive compared to those without sepsis: they cost more, have higher rates of mechanical ventilation, and have longer hospital and ICU lengths of stay.
These results are concordant with previous studies that have identified increased healthcare costs, morbidity, and mortality after a hospitalization with sepsis [9–11]. Recent Ontario data showed that the annual mean healthcare costs for a cohort of individuals who survived a hospitalization with sepsis were significantly higher than a non-sepsis matched cohort, and that these costs remained high, exceeding $20,000 per year, at five years after discharge [2]. Our study adds to these data by demonstrating that these individual costs are among the highest when compared to total population-level health system spending.
Healthcare spending among HCU reflects increased use of inpatient services. A 2016 study examined the types of healthcare costs among top 5% and 1% high-cost users in Ontario. They found that costs related to inpatient health care (acute care, continuing care, or other hospital services) were the largest drivers of healthcare spending among HCU [30]. Among the top 1%, acute hospital costs represented over 30% of total costs, and among the top 1% who did not experience an acute care admission, continuing care represented about half of total costs [30]. Previous studies have demonstrated that readmission rates at 7-, 30-, and 90-days are significantly higher among individuals who experience sepsis during hospitalization compared to those who do not [8, 11, 31]. Recent evidence demonstrates that this increased risk of hospitalization persists for up to 5 years following sepsis (5-year hazard ratio 1.41 [95% CI 1.40–1.43] among those with sepsis, and 1.53 [95% CI 11.50–1.55] among those with severe sepsis) [2]. Sepsis survivors are known to be at a higher risk of death after discharge, and healthcare costs are known to increase before death [32]. It is likely that the costs incurred from hospital-readmission and end-of-life care are contributing to the increased costs among sepsis survivors in our study.
The accurate identification of exposure status can be challenging when using health administrative databases, and this represents a potential limitation in our study. We identified exposure to sepsis via an algorithm that uses ICD-10 codes for infection and organ dysfunction. This algorithm has been validated in ICU and non-ICU populations, with acceptable specificity (> 85%), but with much lower sensitivity (ranging from 25% to 71.9%) [15]. Milder cases of sepsis are known to be under-coded in health administrative databases [15]. This has potential implications for our results. If individuals with milder cases of sepsis experience an increase in healthcare costs after discharge, than this misclassification will bias our results towards the null. However, if milder cases do not experience increased healthcare costs after discharge, then the magnitude of effect due to sepsis that we identified may be inflated. We hope that the new SEPSIS-3 definition and a transition to ICD-11 coding in administrative records will increase the accuracy of sepsis coding and identification in health records.
Previous studies have sought to identify predictors of becoming a HCU. Social determinants of health are strong predictors of being a HCU. [21, 22, 33] Living in a lower income or more marginalized community is associated with increased risk of becoming a HCU [22–24]. Food insecurity has been noted as one of the strongest predictors of being a HCU [24, 34]. The presence of multiple chronic medical conditions is understandably associated with being a HCU [25]. Older age and female sex have also been identified as risk factors [21]. A key strength of this study is that we accounted for these known predictors of future HCU status [24, 35]. We adjusted for common pre-existing comorbidities and neighbourhood level markers of marginalization and income in our propensity score model. We recognise that neighbourhood level markers of social determinants of health need to be handled cautiously, as they do not always perfectly align with individual level income or demographic data [36]. However, these covariates were well balanced after weighting, and our results demonstrate that after using best available methods to control for these known predictors of future healthcare use, sepsis remains a strong independent risk factor for becoming a future HCU.
Conclusion
This study illustrates the impact of sepsis using a well-known marker of healthcare utilization (HCU) that is of interest to health economists and health-system decision-makers. Our results suggest that sepsis can significantly change the trajectory of healthcare utilization among those who were not previously a HCU. These results have important public health and health system implications, and they highlight the economic consequences of sepsis. Programs and policies that aim to prevent sepsis, to reduce the clinical burden of sepsis, and to improve outcomes after sepsis could have significant economic benefit by averting new HCUs. Future research is needed to better predict who is at greatest risk of becoming a HCU after sepsis, and how important social determinants of health modify this risk.
Supplementary Information
Acknowledgements
We would like to thank Dr Alison Fox-Robichaud and Dr Vince Lau, Sepsis Canada representatives, for their comments on an earlier version of this manuscript. This document used data adapted from the Statistics Canada Postal CodeOM Conversion File, which is based on data licensed from Canada Post Corporation, and/or data adapted from the Ontario Ministry of Health Postal Code Conversion File, which contains data copied under license from ©Canada Post Corporation and Statistics Canada. Parts of this material are based on data and/or information compiled and provided by: MOH, Ontario Health, Canadian Institute of Healthcare Information (CIHI). The analyses, conclusions, opinions and statements expressed herein are solely those of the authors and do not reflect those of the funding or data sources; no endorsement is intended or should be inferred. The opinions, results, view, and conclusions reported in this paper are those of the authors and do not necessarily reflect those of OH. No endorsement by OH is intended or should be inferred. Parts of this material are based on data and/or information compiled and provided by CIHI and/or MOH. However, the analyses, conclusions, opinions and statements expressed in the material are those of the author(s), and not necessarily those of CIHI. Census data was adapted from Statistics Canada, Census, 2021. This does not constitute an endorsement by Statistics Canada of this product. Parts of this material are based on data and/or information compiled and provided by Immigration, Refugees and Citizenship Canada (IRCC) current to September 30, 2020. However, the analyses, conclusions, opinions and statements expressed in the material are those of the author(s), and not necessarily those of IRCC. We thank IQVIA Solutions Canada Inc. for use of their Drug Information File. We thank the Toronto Community Health Profiles Partnership for providing access to the Ontario Marginalization Index.
Abbreviations
- HCU
High-cost user
- PHIPA
Personal Health Information Protection Act
- RECORD
Reporting of studies conducted using observational data
- ADG
Aggregated diagnosis groups
- ICU
Intensive care unit
- ECLS
Extracorporeal life support
- IPTW
Inverse probability of treatment weights
- WSD
Weighted standardized difference
Author contributions
All authors contributed significantly to the conception and study design and interpretation of data. HC conducted statistical analyses. KB, FS and VC drafted the manuscript and all authors substantively revised it. All authors have approved the submitted version and have agreed both to be personally accountable for the author's own contributions and to ensure that questions related to the accuracy or integrity of any part of the work, even ones in which the author was not personally involved, are appropriately investigated, resolved, and the resolution documented in the literature.
Funding
This study was supported by ICES, which is funded by an annual grant from the Ontario Ministry of Health (MOH) and the Ministry of Long-Term Care (MLTC). This study also received funding from Sepsis Canada, a Canadian Institute of Healthcare Research (CIHR) Funded National Research Network.
Data Availability
The dataset from this study is held securely in coded form at ICES. While legal data sharing agreements between ICES and data providers (e.g., healthcare organizations and government) prohibit ICES from making the dataset publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS (email: das@ices.on.ca). The full dataset creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to ICES and are therefore either inaccessible or may require modification.
Declarations
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Ethical approval and consent to participate
ICES is an independent, non-profit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyze health care and demographic data, without consent, for health system evaluation and improvement. The use of the data in this project is authorized under Sect. 45 of Ontario’s Personal Health Information Protection Act (PHIPA) and does not require review by a Research Ethics Board.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Rudd KE, Johnson SC, Agesa KM, et al. Global, regional, and national sepsis incidence and mortality, 1990–2017: analysis for the global burden of disease study. Lancet. 2020;395(10219):200–11. 10.1016/S0140-6736(19)32989-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Farrah K, McIntyre L, Doig CJ, et al. Sepsis-associated mortality, resource use, and healthcare costs: a propensity-matched cohort study. Crit Care Med. 2021;49(2):215. 10.1097/CCM.0000000000004777. [DOI] [PubMed] [Google Scholar]
- 3.Sepsis. Accessed 6 July 2023. https://www.who.int/news-room/fact-sheets/detail/sepsis
- 4.Shankar-Hari M, Rubenfeld GD. Understanding long-term outcomes following sepsis: implications and challenges. Curr Infect Dis Rep. 2016;18(11):37. 10.1007/s11908-016-0544-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Iwashyna TJ, Ely EW, Smith DM, Langa KM. Long-term cognitive impairment and functional disability among survivors of severe sepsis. JAMA. 2010;304(16):1787–94. 10.1001/jama.2010.1553. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Prescott HC, Costa DK. Improving long-term outcomes after sepsis. Crit Care Clin. 2018;34(1):175–88. 10.1016/j.ccc.2017.08.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Iwashyna TJ, Ely EW, Smith DM, Langa KM. Long-term cognitive impairment and functional disability among survivors of severe sepsis. JAMA. 2010;304(16):1787–94. 10.1001/jama.2010.1553. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Goodwin AJ, Ford DW. Readmissions among sepsis survivors: risk factors and prevention. Clin Pulm Med. 2018;25(3):79. 10.1097/CPM.0000000000000254. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Prescott HC, Langa KM, Liu V, Escobar GJ, Iwashyna TJ. Increased 1-year healthcare use in survivors of severe sepsis. Am J Respir Crit Care Med. 2014;190(1):62–9. 10.1164/rccm.201403-0471OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Liu V, Lei X, Prescott HC, Kipnis P, Iwashyna TJ, Escobar GJ. Hospital readmission and healthcare utilization following sepsis in community settings. J Hosp Med. 2014;9(8):502–7. 10.1002/jhm.2197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Jones TK, Fuchs BD, Small DS, et al. Post-acute care use and hospital readmission after sepsis. Ann Am Thorac Soc. 2015;12(6):904–13. 10.1513/AnnalsATS.201411-504OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Government of Canada SC. 2016 Census Program. November 16, 2016. Accessed 2 April 2024. 10.25318/1710000901-eng
- 13.Schull MJ, Azimaee M, Marra M, et al. ICES: data, discovery, better health. Int J Popul Data Sci. 2019. 10.23889/ijpds.v4i2.1135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Benchimol EI, Smeeth L, Guttmann A, et al. The reporting of studies conducted using observational routinely-collected health data (record) statement. PLOS Med. 2015;12(10):e1001885. 10.1371/journal.pmed.1001885. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Jolley RJ, Quan H, Jetté N, et al. Validation and optimisation of an ICD-10-coded case definition for sepsis using administrative health data. BMJ Open. 2015;5(12):e009487. 10.1136/bmjopen-2015-009487. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Matheson FI (Unity Health Toronto), Moloney G (Unity Health Toronto), van Ingen T (Public Health Ontario). 2021 Ontario marginalization index: user guide. Toronto, ON: St. Michael’s Hospital (Unity Health Toronto); 2023. Joint publication with Public Health Ontario. https://www.publichealthontario.ca/-/media/Documents/O/2017/on-marg-userguide.pdf
- 17.Lindström AC, Eriksson M, Mårtensson J, Oldner A, Larsson E. Nationwide case–control study of risk factors and outcomes for community-acquired sepsis. Sci Rep. 2021;11(1):15118. 10.1038/s41598-021-94558-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.ACG System Overview. Johns Hopkins ACG® System. Accessed 17 Sept 2024. https://www.hopkinsacg.org/about-the-acg-system/
- 19.Weiner JP, Starfield BH, Steinwachs DM, Mumford LM. Development and application of a population-oriented measure of ambulatory care case-mix. Med Care. 1991;29(5):452–72. 10.1097/00005650-199105000-00006. [DOI] [PubMed] [Google Scholar]
- 20.Starfield B, Weiner J, Mumford L, Steinwachs D. Ambulatory care groups: a categorization of diagnoses for research and management. Health Serv Res. 1991;26(1):53–74. [PMC free article] [PubMed] [Google Scholar]
- 21.Rosella LC, Fitzpatrick T, Wodchis WP, Calzavara A, Manson H, Goel V. High-cost health care users in Ontario, Canada: demographic, socio-economic, and health status characteristics. BMC Health Serv Res. 2014;14(1):532. 10.1186/s12913-014-0532-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Mondor L, Watson T, Kornas K, Bornbaum C, Wodchis WP, Rosella LC. Direct and indirect pathways between low income status and becoming a high-cost health care user in Ontario, Canada: a mediation analysis of health risk behaviors. Ann Epidemiol. 2020;51(28–34): e4. [DOI] [PubMed] [Google Scholar]
- 23.Alberga A, Holder L, Kornas K, Bornbaum C, Rosella L. Effects of behavioural risk factors on high-cost users of healthcare: a population-based study. Can J Public Health Rev Can Santé Publique. 2018;109(4):441–50. 10.17269/s41997-018-0127-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Fitzpatrick T, Rosella LC, Calzavara A, et al. Looking beyond income and education: socioeconomic status gradients among future high-cost users of health care. Am J Prev Med. 2015;49(2):161–71. [DOI] [PubMed] [Google Scholar]
- 25.Wodchis WP, Austin PC, Henry DA. A 3-year study of high-cost users of health care. CMAJ. 2016;188(3):182–8. 10.1503/cmaj.150064. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivar Behav Res. 2011;46(3):399–424. 10.1080/00273171.2011.568786. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med. 2015;34(28):3661–79. 10.1002/sim.6607. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Guilcher SJT, Bronskill SE, Guan J, Wodchis WP. Who are the high-cost users? A method for person-centred attribution of health care spending. PLoS ONE. 2016;11(3): e0149179. 10.1371/journal.pone.0149179. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Wodchis WP, Bushmeneva K, Nikitovic M, McKillop I. Guidelines on Person-Level Costing Using Administrative Databases in Ontario. Published online May 2013. Accessed 5 June 2023. https://tspace.library.utoronto.ca/handle/1807/87373
- 30.Andermann A. Taking action on the social determinants of health in clinical practice: a framework for health professionals. CMAJ. 2016;188(17–18):E474–83. 10.1503/cmaj.160177. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Kim JS, Kim YJ, Ryoo SM, et al. Risk factors for same pathogen sepsis readmission following hospitalization for septic shock. J Clin Med. 2019;8(2):181. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Tanuseputro P, Wodchis WP, Fowler R, et al. The health care cost of dying: a population-based retrospective cohort study of the last year of life in Ontario, Canada. PLoS ONE. 2015;10(3):e0121759. 10.1371/journal.pone.0121759. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.de Oliveira C, Mondor L, Wodchis WP, Rosella LC. Looking beyond administrative health care data: the role of socioeconomic status in predicting future high-cost patients with mental health and addiction. Can J Psychiatry. 2022;67(2):140–52. 10.1177/07067437211004882. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Seth A, Berkowitz MD, Hilary K, Seligman MD, James B, Meigs MD, Sanjay Basu MD. Food Insecurity, Healthcare Utilization, and High Cost: A Longitudinal Cohort Study. 2018; 24. Accessed 2 April 2024. https://www.ajmc.com/view/food-insecurity-healthcare-utilization-and-high-cost-a-longitudinal-cohort-study [PMC free article] [PubMed]
- 35.Rosella LC, Kornas K, Yao Z, et al. Predicting high health care resource utilization in a single-payer public health care system: development and validation of the high resource user population risk tool. Med Care. 2018;56(10):e61–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Buajitti E, Chiodo S, Rosella LC. Agreement between area- and individual-level income measures in a population-based cohort: Implications for population health research. SSM Popul Health. 2020;10: 100553. 10.1016/j.ssmph.2020.100553. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The dataset from this study is held securely in coded form at ICES. While legal data sharing agreements between ICES and data providers (e.g., healthcare organizations and government) prohibit ICES from making the dataset publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS (email: das@ices.on.ca). The full dataset creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to ICES and are therefore either inaccessible or may require modification.