Can this be done by at 8 pm 4/22

1Thavorn K, et al. BMJ Open 2017;7:e017264. doi:10.1136/bmjopen-2017-017264

Open Access

Effect of socio-demographic factors on
the association between multimorbidity
and healthcare costs: a population-
based, retrospective cohort study

Kednapa Thavorn,1,2,3 Colleen J Maxwell,3,4 Andrea Gruneir,3,5,6,7
Susan E Bronskill,3,5 YuQing Bai,3 Anna J Koné Pefoyo,8,9 Yelena Petrosyan,1,5
Walter P Wodchis3,5,10

To cite: Thavorn K,
Maxwell CJ, Gruneir A, et al.
Effect of socio-demographic
factors on the association
between multimorbidity
and healthcare costs: a
population-based, retrospective
cohort study. BMJ Open
2017;7:e017264. doi:10.1136/
bmjopen-2017-017264

► Prepublication history and
additional material for this paper
are available online. To view,
please visit the journal (http://
dx. doi. org/ 10. 1136/ bmjopen-
2017- 017264).

Received 11 April 2017
Revised 3 August 2017
Accepted 11 August 2017

For numbered affiliations see
end of article.

Correspondence to
Dr Walter P Wodchis;
walter. wodchis@ utoronto. ca

Research

AbstrACt
Objectives To estimate the attributable costs of
multimorbidity and assess whether the association
between the level of multimorbidity and health system
costs varies by socio-demographic factors in young (<65 years) and older (≥65 years) adults living in Ontario, Canada. Design A population-based, retrospective cohort study setting The province of Ontario, Canada Participants 6 639 089 Ontarians who were diagnosed with at least one of 16 selected medical conditions on 1 April 2009. Main outcome measures From the perspective of the publicly funded healthcare system, total annual healthcare costs were derived from linked provincial health administrative databases using a person-level costing method. We used generalised linear models to examine the association between the level of multimorbidity and healthcare costs and the extent to which socio- demographic variables modified this association. results Attributable total costs of multimorbidity ranged from C$377 to C$2073 for young individuals and C$1026 to C$3831 for older adults. The association between the degree of multimorbidity and healthcare costs was significantly modified by age (p<0.001), sex (p<0.001) and neighbourhood income (p<0.001) in both age groups, and the positive association between healthcare costs and levels of multimorbidity was statistically stronger for older than younger adults. For individuals aged 65 years or younger, the increase in healthcare costs was more gradual in women than in their male counterparts, however, for those aged 65 years or older, the increase in healthcare costs was significantly greater among women than men. Lastly, we also observed that the positive association between the level of multimorbidity and healthcare costs was significantly greater at higher levels of marginalisation. Conclusion Socio-demographic factors are important effect modifiers of the relationship between multimorbidity and healthcare costs and should therefore be considered in any discussion of the implementation of healthcare policies and the organisation of healthcare services aimed at controlling healthcare costs associated with multimorbidity. bACkgrOunD Multimorbidity, the presence of two or more coexisting conditions within a single person, is increasingly prevalent due to advances in life-extending medical treatments and increases in life expectancy.1 2 Internationally, the prevalence of multimorbidity has been shown to range from 17% in young adults3 to 82% in older adults living in nursing homes.4 In the province of Ontario, Canada, the prev- alence of multimorbidity based on 16 selected conditions rose from 17.4% in 2003 to 24.3% in 2009, and this increase was evident across all age groups.5 Higher levels of multimorbidity are asso- ciated with impaired physical functioning,6 strengths and limitations of this study ► This population-based study was based on a large sample size and used robust costing and generalised linear model regression techniques. ► The availability of linked and patient-level health administrative databases allows the estimation of the total health system costs associated with multimorbidity from all healthcare sectors. ► The use of health administrative databases can also minimise potential recall and non-response biases that are commonly found in survey data. ► The total healthcare costs reported in this study may be underestimated because they were derived based on 16 selected medical conditions. Moreover, it was not possible to measure certain costs (eg, deductibles and copayments borne by supplemental health insurance, out-of-pocket beneficiary payments and indirect costs associated with caregiving) with our data. ► The study did not take into account particular clusters of medical conditions. It is possible that the relationship between multimorbidity and healthcare costs may vary according to the types and patterns of comorbid medical conditions. http://bmjopen.bmj.com/ http://dx.doi.org/10.1136/bmjopen-2017-017264 http://dx.doi.org/10.1136/bmjopen-2017-017264 http://dx.doi.org/10.1136/bmjopen-2017-017264 http://crossmark.crossref.org 2 Thavorn K, et al. BMJ Open 2017;7:e017264. doi:10.1136/bmjopen-2017-017264 Open Access poorer quality of life,7 more frequent use of health services and higher risk of death.8 In addition, individ- uals with multimorbidity may experience faster disease progression and require more complex medical care.9 Consequently, these individuals may be at a higher risk of receiving suboptimal care,10 inappropriate prescrip- tions11 and experiencing potentially preventable hospital- isations.12 These adverse health outcomes can impose a substantial burden on patients, family caregivers and the healthcare system. The relationship between multimorbidity and health- care costs is well-documented and has been shown to be curvilinear or exponential across jurisdictions. The average Medicare payments in the USA ranged from US$1154 among part A and part B beneficiaries with one chronic condition to US$13 973 among beneficiaries with at least four chronic conditions (a 12-fold difference).12 Similarly, the mean total health system costs among older adults with multimorbidity in Switzerland were nearly six times higher than among those without multimorbidity.13 Despite an abundance of research describing the rela- tionship between multimorbidity and healthcare costs, existing studies have some important methodological and conceptual limitations. Some previous studies14 15 used ordinary least squares (OLS) regression despite the fact that the positively skewed distribution of cost data often violates the normality assumption of OLS.16 Others attempted to overcome this problem by transforming cost data to the logarithmic scale13 17; however, this trans- formation may still result in interpretation problems, as regression on transformed costs provides the predic- tion of a median instead of the arithmetic mean costs.18 Importantly, the role of socio-demographic characteris- tics as effect modifiers of the relationship between multi- morbidity and healthcare costs remains poorly described, although previous research has shown that the specific types of disease clusters vary by age and sex2 19 and that multimorbidity is more prominent in selected visible minority and low-socioeconomic status populations.20 The objectives of this study were, therefore, to estimate the health system costs attributable to multimorbidity using a more rigorous and appropriate approach, and to assess the extent to which the relationship between the level of multimorbidity and health system costs varies according to socio-demographic characteristics. MethODs study design and sample This population-based, retrospective cohort study included all residents of the province of Ontario between 1 April 2001 and 31 March 2010, who were enrolled in the Ontario Health Insurance Plan (OHIP), and were diagnosed with at least one of the following selected 16 medical conditions between 1 April 2001 and 31 March 2009 (study index date): acute myocardial infarction (AMI), arthritis, asthma, cancer, cardiac arrhythmia, chronic coronary syndrome, chronic obstructive pulmonary disorder (COPD), congestive heart failure, dementia, depression, diabetes, hypertension, osteopo- rosis, renal failure, rheumatoid arthritis and stroke. These conditions were selected because previous research and clinical experts agreed that they were highly prevalent and represented a substantial care and economic burden for Canada’s healthcare system.5 21 We excluded indi- viduals if they met the following criteria: had an invalid health card number, were older than 105 years, died or moved out of the province prior to the index date. Indi- viduals with no contact with the healthcare system within the past 5 years prior to the index date were also excluded (excepting infants), as they may have left the province or experienced an unreported death. Data sources We linked multiple provincial health administrative data- bases anonymously using unique encrypted identifiers. The Discharge Abstract Database provides data for all hospital discharges in Ontario, and the OHIP claims database includes billing claims for all physician encoun- ters. We used the Registered Persons Database to iden- tify Ontarians who were eligible for health insurance coverage and derive their age. The linked database was housed and secured at the Institute for Clinical Evaluative Sciences (ICES) under data security and privacy policies and procedures approved by the Information and Privacy Commissioner of Ontario. This study was approved by the Research Ethics Board at Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada. Each medical condition was defined using diagnostic algorithms and consultation codes that have been vali- dated or used in previous studies. We defined six condi- tions (AMI, asthma, chronic heart failure, COPD, diabetes and hypertension) based on validated population-derived registries held at ICES.22–28 These conditions were all defined based on one diagnosis recorded in acute care or two diagnoses recorded in ambulatory care (physician) records within a 2-year period (ie, between 2007/2008 and 2008/2009), except for AMI, which was defined using acute care records in 2008/2009. A similar approach was adopted to define the remaining medical conditions including arthritis, cancer, cardiac arrhythmia, chronic coronary syndrome, dementia, depression, osteoporosis, renal failure, rheumatoid arthritis and stroke. A list of diagnostic codes used to define these medical conditions are shown in online supplementary appendix 1. Measures Healthcare costs Healthcare costs were estimated from the perspective of the publicly funded healthcare system; accordingly, only direct costs borne to the Ontario Ministry of Health and Long-Term Care were considered. In Ontario, medically necessary hospital and physician services are paid for by the publicly financed health insurance plan; however, public coverage for prescription drugs is primarily limited to residents aged 65 years and over, social assistance https://dx.doi.org/10.1136/bmjopen-2017-017264 3Thavorn K, et al. BMJ Open 2017;7:e017264. doi:10.1136/bmjopen-2017-017264 Open Access recipients as well as those with high prescription drug costs compared with their net household income. We identified, measured and valued direct healthcare costs by applying a person-level costing technique that was developed and validated based on the Ontario health administrative data.29 We calculated the costs of inpatient hospitalisations, emergency department visits, same day surgeries and inpatient rehabilitation by multiplying the weighted volume of services by the average provincial costs per weighted case. We obtained the costs of fee-for-ser- vice physician and outpatient diagnostic or laboratory services through OHIP fee approved as outlined in the Ontario Health Insurance Schedule of Benefits and Fees.30 Non-fee-for-service physician payments were calcu- lated by applying applicable capitation payments or the median amount reimbursed for the same service code for the specific fiscal year.29 Costs for high-cost medical device equipment were estimated from the amount reimbursed to patients recorded in the Assistive Devices Program database. Complex continuing care and inpa- tient psychiatric costs were based on case mix, number of days in care and resource utilisation groups.31 Patient costs for long-term care were estimated based on a fixed per diem according to prevailing government payment rates, and costs for home care were estimated using the average cost per hour. We used pharmacy payments recorded in the Ontario Drug Benefit database to capture prescription medication costs for individuals eligible for public coverage. Annual total direct healthcare costs were the sum of costs across healthcare sectors for each patient for a 1-year period after the study index date, that is, from April 2009 to March 2010. We categorised healthcare costs into five components: physician, hospital, drug, continuing care and other healthcare delivery costs. Physician costs included profes- sional fees paid by the provincial insurance plan directly to physicians in private practice. Hospital costs included amounts paid to healthcare institutions, including those providing acute care, extended and chronic care, rehabil- itation and convalescent care, psychiatric care as well as drugs dispensed in hospitals. Drug costs consisted of the costs of prescriptions dispensed at outpatient pharma- cies to individuals eligible for provincial coverage while continuing care costs included expenditure on home care and residential long-term (nursing-home) care. The other healthcare delivery costs category represented expenditures on an assistive device programme that subsi- dises high-cost equipment, such as wheel chairs, walkers, continuous positive airway pressure devices and insulin pumps, for patients with physical disabilities. All costs were expressed in 2009 Canadian Dollars (C$). Independent variables Multimorbidity was defined as the occurrence of two or more chronic diseases among the 16 selected conditions within a single individual and was categorised into five groups. A categorical variable was created to capture those with no multimorbidity (single disease only), two, three, four and five or more multimorbid conditions. Socio-demographic variables included age, sex, income and level of marginalisation. As prescription drug costs among Ontarians aged <65 years were primarily covered by private drug plans, we ran separate regressions for younger (<65 years) and older (65+ years) cohorts, and also included a continuous variable for age in the models. Income level was categorised into five quintiles, with the lowest quintile reflecting the lowest income level. We used the Ontario Marginalisation Index, a validated census- based and geographically based index, as a proxy for individual-level socio-demographic factors.32 The index consisted of four dimensions of marginalisation: mate- rial deprivation; residential instability; ethnic concentra- tion and dependency. Lower scores on each dimension represent areas that are the least marginalised and higher scores represent areas that are the most marginalised. This index has been shown to be associated with several health outcomes.33 We also controlled for other factors that may confound the impact of multimorbidity on healthcare costs, such as the type of primary care model and geographic location. Selection of such factors was guided by previous health- care cost studies12 34 35 and was subject to the availability of data on these factors in Ontario administrative data- bases. The payment scheme of primary care services was categorised into three groups: group-based teams with capitation/salary and team-based payment (family health teams/other group models); capitation or blended payment models (family health networks/family health organisations) or primarily fee for service (family health groups and non-rostered patients). Lastly, we assigned a geographic location to each individual using the Rurality Index for Ontario,36 whereby a value >40 was considered
to be a designated rural area.

Analysis
Annual healthcare costs per capita were described by
health service sector, age group (<65 vs ≥65 years), the degree of multimorbidity and each of the independent factors, such as sex, age group and level of marginalisa- tion. Multivariate regression analyses were used to assess the incremental costs of interest in this study. To iden- tify the regression model that best fits the cost data, we followed the steps suggested by Manning and Mullahy.37 We first ran OLS of the logarithmic transformation of cost data on the number of medical conditions and other confounding factors; however, the OLS regression was deemed inappropriate because the residuals were not normally distributed. Therefore, the generalised linear model (GLM) with a log-link function and a gamma distri- bution was chosen because a modified Park test suggested that the variance was proportional to the conditional mean. The GLM allows us to estimate mean healthcare costs without the need for retransformation. Attributable costs due to multimorbidity were esti- mated by subtracting the mean predicted cost of one medical condition from the mean predicted cost of two 4 Thavorn K, et al. BMJ Open 2017;7:e017264. doi:10.1136/bmjopen-2017-017264 Open Access conditions, two from three conditions, three from four conditions and four from at least five conditions, respec- tively. To investigate whether the relationship between the level of multimorbidity and healthcare costs was moderated by socio-demographic factors, we added two-way interaction terms between the level of multimor- bidity and each socio-demographic factor, including sex, age, income level, deprivation quintile, instability quin- tile, dependency quintile and ethnic concentration quin- tile. The significance of interaction terms was assessed by comparing the likelihood ratio of the full model with all interaction terms with the model without interaction terms using the likelihood ratio test. The model performance, including goodness of fit and specifications, was examined by checking the scaled deviance, Pearson’s χ2 statistics and residual plots, respec- tively. All analyses were performed using SAS statistical software for UNIX (V.9.3; SAS Institute, Cary, North Carolina, USA). results We identified a cohort of 6 639 089 individuals living with at least one of the selected 16 medical conditions in Ontario in 2009 (see online supplementary appendix 2 for baseline characteristics). Our cohort represents about 50% of the total population in the province of Ontario in 2009. Close to half of the study cohort (48%) had at least two selected medical conditions, and this prevalence was found to increase with age. The majority of the study cohort was younger than 65 years of age (75%) and just over half was female (53%). Nearly all individuals (91%) resided in non-rural areas, and about one-third (33%) lived in neighbourhoods with a high proportion of diverse ethnic groups. The total annual healthcare cost estimated for the study cohort was C$26.5 billion. As shown in figure 1, individuals living with at least two selected medical conditions represented 24.4% of the total population of Ontario (~13 million) but accounted for approximately two-thirds (67.9%) of total allocatable healthcare costs in 2009/2010. By contrast, individuals without multimor- bidity who accounted for 76% of the total population were responsible for only 32.1% of total allocatable healthcare costs. On average, annual total costs per capita amounted to C$2217 in individuals <65 years and C$9398 in those aged 65 years or older. Table 1 shows the annual total costs per capita by base- line characteristics for young and older adults. For both age groups, per capita total healthcare costs were higher in women than in men. The average healthcare costs increased with older age, and greater levels of marginali- sation were associated with higher healthcare costs in both age groups. Mean total healthcare costs were the highest among individuals living in the most deprived and most unstable areas as well as those who were highly depen- dent; however, mean total costs decreased as income level increased. Figure 2 illustrates the distribution of total cost per capita by type of services. Among individuals <65 years, hospitalisation was the primary cost driver and respon- sible for 47% of total healthcare costs, followed by physi- cian costs (32%), drug costs (10%) and continuing care costs (6%). For older adults, hospital costs remained the largest cost component (41%), followed by continuing care costs (23%), drug costs (19%) and physician costs (15%). Figure 2 also reveals that unadjusted mean total costs increased with additional numbers of medical conditions, ranging from C$1352 in individuals <65 years without multimorbidity to C$13 105 in those living with Figure 1 Distribution of total number of population and total health system costs in Ontario from 1 April 2009 to 31 March 2010. https://dx.doi.org/10.1136/bmjopen-2017-017264 5Thavorn K, et al. BMJ Open 2017;7:e017264. doi:10.1136/bmjopen-2017-017264 Open Access Table 1 Annual per capita healthcare costs by baseline characteristics and age group, 1 April 2009 to 31 March 2010 <65 years (n=5 004 699) ≥65 years (n=1 634 390) N Per capita healthcare cost (C$) N Per capita healthcare cost (C$) Mean (SD) Median (IQR) Mean (SD) Median (IQR) All cohort 5 004 699 2217 (9630) 502 (193–1317) 1 634 390 9398 (19 796) 2982 (1448–7178) Sex Female 2 618 591 2311 (9044) 624 (248–1546) 923 053 9526.96 (19 245) 2991.97 (1461–7344) Male 2 386 108 2113 (10 233) 378.67 (132–1058) 711 337 9230.31 (20 488) 2968.13 (1431–6982) Age group (years) <20 809 782 997 (6420) 257 (103–600) 20–44 1 784 314 1835 (7997) 440 (155–1171) 45–64 3 247 243 2910 (11 414) 684 (291–1725) 65–74 1 219 877 6424 (16 464) 2363 (1173–4757) 75+ 797 750 12 517 (22 351) 3964 (1884–12 277) Income quintile Lowest 935 048 2822 (11 333) 580 (206–1699) 314 616 10 646 (21 501) 3325 (1596–8667) Middle-low 970 797 2360 (10 276) 521 (199–1380) 336 928 9529 (20 218) 3053 (1501–7296) Middle 999 087 2107 (9146) 498 (195–1268) 318 557 9319 (19 552) 2992 (1470–7114) Middle-high 1 042 284 2008 (8 899) 487 (195–1226) 322 798 9120 (19 279) 2916 (1426–6873) Highest 1 009 890 1903 (8391) 475 (192–1180) 331 022 8549 (18 309) 2747 (1351–6352) Rurality index Non-rural 4 579 691 2206 (9 605) 509 (197–1320) 1 459 014 9448 (19 998) 3005 (1470–7161) Rural 356 361 2522 (10 112) 501 (197–1441) 157 864 9333 (18 303) 2918 (1400–7798) Deprivation quintile Least deprived 1 282 898 1894.17 (8 596 59) 476 (193–1170) 371 547 9167 (19 628) 2823 (1380–6709) Less deprived 1 136 731 2015 (8 810) 489 (196–1231) 368 124 8935 (18 928) 2898 (1423–6759) Somewhat deprived 982 133 2193 (9 240) 504 (196–1311) 346 326 9165 (19 300) 2978 (1463–7030) Very deprived 808 152 2438 (10 281) 511 (200–1443) 293 434 9541 (19 951) 3100 (1520–7467) Most deprived 705 593 2941 (11 861) 600 (210–179) 228 501 10 517 (21 250) 3326 (1599–8570) Instability quintile Least dependent 1 211 734 2007 (8 674) 489 (188–1250) 188 787 8149 (19 413) 2713 (1307–5882) Less dependent 1 179 936 2078 (9 134) 500 (195–1275) 276 819 8359 (18 652) 2777 (1353–6167) Somewhat dependent 976 538 2230 (9 793) 506 (198–1320) 303 853 8717 (19 018) 2849 (1401–6548) Very dependent 808 196 2349 (9 954) 515 (201–1375) 326 662 9068 (19 195) 2944 (1458–6958) Continued 6 Thavorn K, et al. BMJ Open 2017;7:e017264. doi:10.1136/bmjopen-2017-017264 Open Access five or more medical conditions, corresponding to a 10-fold increase. On the other hand, while C$4185 was spent on older adults without multimorbidity, spending increased by about fivefold to C$19 196 in those living with five or more medical conditions. Table 2 shows adjusted attributable costs of multimor- bidity after controlling for other factors. Among individ- uals <65 years, the attributable total cost was C$377 in those living with two medical conditions and C$2073 in those living with at least five medical conditions, corre- sponding to a sixfold increase in attributable cost. Simi- larly, attributable total costs in older adults also rose with increasing number of medical conditions, ranging from C$1026 in those with two medical conditions to C$3831 in those with five or more. The magnitude of an incre- mental cost, however, depended on the reference cate- gory. Specifically, one additional medical condition in young adults without multimorbidity led to an attribut- able cost of C$377, while for young adults who already had three medical conditions, one additional health condition resulted in a total cost of C$798. These incre- mental costs were even greater in older adults, among whom the incremental cost rose from C$1026 (one vs two conditions) to C$1652 (three vs four conditions). Similar patterns were observed for subdivided healthcare costs, which varied across age groups (table 2). An additional <65 years (n=5 004 699) ≥65 years (n=1 634 390) N Per capita healthcare cost (C$) N Per capita healthcare cost (C$) Mean (SD) Median (IQR) Mean (SD) Median (IQR) Most dependent 739 103 2650 (10 947) 550 (213–1507) 511 811 10 961 (20 953) 3381 (1636–9336) Ethnic concentration quintile Lowest 564 476 2398 (9 766) 500 (200–1370) 283 980 9309 (18 529) 2983 (1463–7533) Middle-low 756 120 2288 (9 552) 491 (196–1317) 304 526 9170 (18 773) 2969 (1458–7283) Middle 854 573 2280 (9 780) 497 (196–1317) 305 524 9540 (19 678) 3011 (1478–7419) Middle-high 1 028 876 2190 (9 565) 502 (195–1309) 294 164 9600 (20 240) 3012 (1473–7266) Highest 1 711 462 2124 (9 468) 528 (199–1331) 419 738 9288 (20 751) 2981 (1441–6694) Table 1 Continued Figure 2 Unadjusted mean total healthcare cost per capita for Ontario adults, by service type, number of conditions and age group from 1 April 2009 to 31 March 2010. 7Thavorn K, et al. BMJ Open 2017;7:e017264. doi:10.1136/bmjopen-2017-017264 Open Access medical condition caused a onefold to threefold increase in the costs of each health sector except for hospital care, for which incremental costs increased steadily from C$185 to C$802 in the younger cohort and C$232 to C$1060 in the older adult cohort. We also found that the association between the number of medical conditions (ie, the degree of multimorbidity) and healthcare costs was significantly modified by age and sex for both young and older adults (table 3), and the positive association between healthcare costs and levels of multimorbidity was significantly stronger for older than younger adults. For individuals aged 65 years or younger, the increase in healthcare costs was more gradual in women than in their male counterparts, however, among those aged 65 years or older, the increase in healthcare costs in women was significantly greater than in men. For both age groups, we observed small interaction effects between the number of medical conditions and other socio-demographic factors. The rise in healthcare costs as the level of multimorbidity increased was less pronounced among high-income individuals than low-in- come individuals, and the association between the level of multimorbidity and healthcare costs was significantly modified by the level of deprivation, instability, depen- dency and ethnic concentration. The positive …