The role of mental health and addiction among high-cost patients: a population-based study

Abstract

Aims: Previous work found that, among high-cost patients, those with a majority of mental health and addiction (MHA)-related costs (>50%) incur over 30% more costs than other high-cost patients. However, this work did not examine other high-cost patients in depth or whether they had any MHA-related costs. The objective of this analysis was to examine the role of MHA-related care among other high-cost patients.

Methods: Using administrative healthcare data from Ontario, Canada, this study selected all patients in the 90th percentile of the cost distribution in 2012. It focused primarily on two groups based on the percentage of MHA-related costs relative to total costs: (1) high-cost patients with some MHA-related costs (0% > and <50%) and (2) high-cost patients with no MHA-related costs (0%). We examined socio-demographic and clinical characteristics, utilization and costs for both groups, and modeled patient-level costs using appropriate regression techniques. We also compared these groups with high-cost patients with a majority of MHA-related costs (>50%).

Results: High-cost patients with some MHA-related costs incurred over 40% more costs than those without ($27,883 vs $19,702). Patients with some MHA-related costs were older, lived in poorer neighborhoods, and had higher levels of comorbidity compared to those without. After controlling for relevant variables, having any type of MHA-related utilization increased costs by $2,698. Having a diagnosis of psychosis had a large impact on costs.

Limitations: This study did not examine children and adolescents. We were only able to account for 91% of all costs incurred by the public third-party payer; addiction-related costs from community-based agencies were not available.

Conclusions: High-cost patients with MHA incur higher costs compared to those without. When considering interventions aimed at high-cost patients, policy-makers should consider their complex nature, specifically both their physical and MHA-related comorbidities.

Keywords:

• High-cost patients
• mental health
• addiction
• healthcare costs
• administrative data

Introduction

The literature has shown that a small proportion of healthcare users—high-cost patients—account for a disproportionately large share of healthcare costs^1–3. In recent years, there has been increasing interest in these patients by policy-makers, namely around how best to reduce costs while simultaneously improving patients’ health and care experiences, known as the “triple aim”^4–6. According to Weil⁷, “If we can better care for those with the greatest needs, we can improve their lives, and if through prevention, coordination, or attention to social needs we can avert the need for even a small portion of their care, we can save money as well” (7).

Most work on high-cost patients has examined these patients in their entirety; little research has explored patient sub-groups, although exceptions exist⁶. Furthermore, few studies have examined the role of mental health and addiction (MHA) among the high-cost population. Patient profiles, needs, and utilization patterns likely vary between sub-populations⁶. For example, as noted by Forget et al.⁸, “not all of the high-cost users are elderly” (e151). Thus, an accurate understanding of who constitutes a high-cost patient and how they utilize the healthcare system is required as one model of care does not “fit all. To optimize investments in this area, we need a better understanding of the high-cost population, namely whether there are any patient sub-groups; this may help inform the design of appropriate interventions⁶.

Previous work done by our group found that, among all high-cost patients in Ontario, mental health high-cost patients—those with a majority of MHA-related costs—incur over 30% more costs than other high-cost patients⁹. These high-cost patients constitute a particular segment of the patient population, many of which have a diagnosis of psychosis or mood disorders, with the bulk of their costs made up of psychiatric hospitalizations. However, our previous work did not examine other high-cost patients in detail or the role of MHA among these patients. We posit that patients with mental illnesses and addictions have higher costs of care than those without these disorders. Prior research suggests that MHA can play an important role, even among high-cost patients without severe mental illness¹⁰. For example, one study in the US found that, among the top 5% of Medicare beneficiaries, 23.6% had a mental disorder in 2004¹. A more recent study from the US, which examined patient sub-groups by payer category, found high prevalence of mental health disorders among the top 1% of Medicaid beneficiaries⁶. In particular, they found that one quarter of patients had been diagnosed with depression, another quarter with anxiety, and almost one fifth with bipolar disorder. However, these studies did not specifically examine the role of MHA among high-cost patients or provide an in-depth understanding of how MHA-related care affects these patients’ healthcare utilization and costs^11,12. Furthermore, most work has been done in the US^1,6,11,12 and has focused on specific sub-populations covered under public healthcare insurance plans only, such as Medicare¹ and Medicaid¹², which may limit generalizability of findings to the entire population.

The objective of this analysis was to examine the role of MHA-related healthcare on total direct healthcare costs for a population-based sample of high-cost patients in Ontario, Canada, for 2012. In particular, we are interested in understanding whether the proportion of MHA-related costs has an effect on total costs. We also provide some suggestions on what interventions should account for to address the heterogeneous needs of this population.

Methods

Data

We used administrative healthcare data available from the Institute for Clinical Evaluative Sciences, in Toronto, Ontario. This data repository contains individual-level linkable and longitudinal data on most publicly funded healthcare services for all legal residents of Ontario, Canada’s largest province. It includes several health services databases, many of which have been validated and described in the literature¹³, and used as a source for costing analyses in Ontario^9,14,15. It currently does not include data on addiction-related healthcare provided by community-based agencies. A full description of each database can be found in Supplementary Table 1.

We examined data for all adult patients (aged 18 and older) who had at least one encounter with the Ontario healthcare system from April 2012 to March 2013, after excluding all individuals (16%) who did not have a valid health card number and, thus, did not have data on healthcare utilization. We defined high-cost patients as those in the 90th percentile of the cost distribution, in line with previous work^9,12. This threshold enabled us to select a larger cohort of patients compared to other thresholds used.

Previous work examined high-cost patients with majority MHA costs, which were defined as those patients for whom costs related to MHA care accounted for 50% or more of their total healthcare costs⁹. Defining high-cost patients with majority MHA costs based on specific diagnoses can be challenging when using administrative healthcare data. In Ontario, physician billing codes are not diagnosis-specific (with the exception of psychosis) and may be used for several mental disorders (diagnosis-specific codes are only available for hospitalization and emergency department visit data). This is less of a concern for diseases, such as cancer or kidney disease, where there are provincial registries and diagnosis identification is straight-forward. Furthermore, defining high-cost patients with majority MHA costs as those with severe mental illness only would have excluded about a third of high-cost patients who do not require being hospitalized, but have high costs with outpatient drugs and physician services. Thus, we decided to apply the definition used in previous published research⁹. This work found that changing the threshold to 60% had little impact on the cohort definition, as mental health-related costs accounted for most (87.4%) of total costs of this group. Although these patients are not the primary focus of this analysis, we included descriptions on these patients for the sake of comparability.

In addition to this patient group, we defined two other patient groups of particular interest: (1) high-cost patients with some MHA-related costs, i.e. those for whom costs related to MHA care accounted for less than 50% of their total healthcare costs (but higher than zero); and (2) high-cost patients with no MHA-related costs, i.e. those with no costs related to MHA utilization. We divided our cohort into these patient groups to test the hypothesis of whether having MHA-related care leads to higher average costs of care. The idea is that having a mental disorder and/or addition likely increases patient complexity, and, thus, leads to higher healthcare utilization and costs.

The administrative healthcare databases include separate data on utilization and individual provider and corporate unit cost information collected by the Ontario Ministry of Health and Long-term Care. Therefore, we used a cost estimation algorithm, available at the Institute for Clinical Evaluative Sciences, to estimate all patient-level total healthcare costs borne by the public third-party payer only (i.e. the Ontario Ministry of Health and Long-term Care)¹³. These included costs with inpatient hospitalizations (both acute and psychiatric hospitalizations), emergency department (ED) visits and other ambulatory care, physician-related visits and lab tests, outpatient prescription drugs (for individuals covered under the public provincial drug plan¹⁶), rehabilitation, home care, complex continuing care, long-term care, and assistive devices. All costs were reported in 2012 Canadian dollars.

In line with previous work, MHA-related costs were defined as all costs associated with psychiatric hospitalizations (i.e. hospitalizations with any DSM-IV code and/or any ICD-10 F00-F99 code), emergency department visits for MHA-related reasons or self-harm (i.e. emergency department visits with any ICD-10 F00-F99 or any X60–X84 code), MHA-related physician and outpatient services, prescription drugs used to treat psychiatric disorders (i.e. antidepressants, anti-psychotics and mood stabilizers), and home care psychology services⁹. All remaining costs described above were defined as non-MHA-related costs⁹.

Analysis

We compared the three high-cost groups on variables likely to influence healthcare utilization and, thus, costs. These included socio-demographic characteristics such as sex; age; neighborhood income quintile at the census tract level; immigrant/refugee status, obtained from data collected by Immigration, Refugees and Citizenship Canada; long-term care residence and urban/rural residence. We also examined the number of comorbidities measured by the number of Aggregate Diagnosis Groups (ADGs), which take into account the duration, severity and etiology of the condition; diagnostic certainty (i.e. how well the diagnosis was established), and specialty care involvement¹⁷. We estimated and compared overall and average costs for major cost categories (hospitalizations, physician services, ED visits, prescription drugs, home care, long-term care, and other care) for each patient group.

We also used regression analysis to examine the explanatory power of each variable described beforehand on healthcare costs, but only for high-cost patients with majority MHA-related costs. The generalized linear model has become the preferred strategy for modeling healthcare costs^18–22; the main advantage is that predictions are made on the raw cost scale, and, thus, no retransformation is required. Furthermore, it allows for heteroskedasticity through the choice of the distributional family²³. We performed the modified Park test²⁰ to determine the most appropriate distribution of the dependent variable; the results suggested the Poisson distribution. Given the severe skewness in healthcare costs, we chose the log link where variables act multiplicatively on the mean.

We ran two regression models (Models 1 and 2). The dependent variable for both models was the total cost for each patient in 2012. Given that some patients had missing data on key explanatory variables, such as neighborhood income quintile (n = 4,576), all models were run on the sample of high-cost patients with complete data only. We found that patients with missing data did not differ from those with complete data.

Our main independent variable in Model 1 was a binary outcome indicating whether the patient had any MHA-related costs. Our control variables included sex, age, age² (to control for non-linear effects in age), neighborhood income quintile, immigration/refugee status, rurality indicator, long-term care indicator, number of comorbidities, and administrative health region dummies (in Ontario, these are called Local Health Integration Networks; they are responsible for the regional administration of public healthcare services).

Our main independent variables in Model 2 were dummy diagnosis variables indicating whether the patient was hospitalized for a particular chronic condition. We considered the chronic condition associated with the main diagnosis (for example, if a patient had been hospitalized for COPD and depression, where the first was the main diagnosis and the second was listed as a secondary diagnosis, we classified the encounter as a COPD hospitalization). We focused on hospitalizations, since these are the most costly healthcare encounters among high-cost patients^3,9,10. Furthermore, we chose the most common (physical and mental) chronic conditions typically found among high-cost patients, as suggested in the literature^1,24; these included cancer; diabetes; ischemic heart disease; heart failure; stroke; COPD; renal failure; schizophrenia or other psychoses; mood disorders; dementia, and substance abuse (see Supplementary Table 2 for codes). In addition, we included a dummy variable (based on hospital diagnoses) indicating the presence of both a mental and physical health condition to capture patient complexity (we excluded ICD-10 F8-F9 codes, which tend to be more prevalent among children and youth, and ICD-10 codes O, P, and Q, which correspond to pregnancy, childbirth, and the puerperium; conditions originating in the perinatal period; and congenital malformations, deformations, and chromosomal abnormalities, respectively). We used the same control variables included in Model 1.

For both models, we obtained values for the coefficients and the marginal effects, but only report the latter, as these are more useful for interpretation. Robust standard errors were estimated for all models.

Sensitivity analyses

We defined high-cost patients as those in the 90th percentile of the cost distribution. However, other work has defined high-cost patients as those in the 95th or 99th percentiles^2,25. Thus, we also examined these cut-off points to test the sensitivity of our results in both our descriptive and regression-based analyses.

We also ran a data-related robustness check. Outpatient drugs account for a large part of mental healthcare spending by payers²⁶. The provincial public drug plan (Ontario Drug Benefit Program), for which we have data on, provides drug coverage for seniors (people aged 65 and older) and individuals younger than 65 years on social assistance only¹⁶. As a result, we did not have drug data for roughly 22% of our high-cost population. To address this potential bias, we carried out all analyses for our original cohorts excluding drug costs for all patients younger than 65, in line with previous work⁹.

Results

Patient selection

In 2012, there were 987,887 high-cost patients, which cost the Ontario healthcare system $23.8 billion (Table 1). The average cost per high-cost patient was $24,094. (Dollar amounts throughout this article are in 2012 Canadian dollars, unless otherwise specified. As of late September 2017, one Canadian dollar was worth ∼0.80 US dollars.) Among these patients, there were 51,457 with a majority of MHA-related costs, 455,479 patients with some MHA-related costs, and 480,951 with no MHA-related costs. There was a clear cost gradient among the three groups where average costs decreased with the proportion of MHA-related costs relative to overall costs—$31,611, $27,883, and $19,702 (Table 1). Excluding high-cost patients with a majority of MHA-related costs, we found that high-cost patients with some MHA-related costs incurred 40% higher costs those with no MHA-related costs (Table 1). Furthermore, total costs of care for high-cost patients with some MHA-related costs were larger than those without MHA costs ($12.7 billion vs $9.5 billion, respectively). Results were qualitatively similar for patients in the 95th percentile ($42,680 and $35,312). For those in the 99th percentile, patients with no MHA costs had higher mean costs than those with some MHA costs ($88,119 and $97,281); this was mainly due to higher costs of hospitalizations and other institution-based care, dialysis care, and cancer care among this group) (results available upon request).

Table 1. Socio-demographic and clinical characteristics of high-cost patient groups in Ontario, 2012.

Characteristic	High-cost patients with majority mental health and addiction-related costs		High-cost patients with some mental health and addiction-related costs		High-cost patients with no mental health and addiction-related costs		All high-cost patients
	n	%	n	%	n	%	n	%
Sample	51,457	5	455,479	46	480,951	49	987,887	100
Cost (2012 Canadian $)
Total	1.6 billion	7	12.7 billion	53	9.5 billion	40	23.8 billion	100
Average	31,611	—^a	27,883	—^a	19,702	—^a	24,094	—^a
Sex
Female	26,175	50.9	281,040	61.7	263,108	54.7	570,323	57.7
Male	25,282	49.1	174,439	38.3	217,843	45.3	417,564	42.3
Age (years)^b
Less than 30	10,048	19.5	20,775	4.6	43,897	9.1	74,720	7.6
30–39	9,021	17.5	26,649	5.9	51,813	10.8	87,483	8.9
40–49	10,899	21.2	38,142	8.4	45,837	9.5	94,878	9.6
50–59	10,690	20.8	55,806	12.3	68,297	14.2	134,793	13.6
60–69	5,433	10.6	80,130	17.6	89,680	18.6	175,243	19.3
70–79	2,732	5.3	99,125	21.8	89,799	18.7	191,656	19.1
80 or more	2,634	5.1	134,852	29.6	91,628	19.1	229,114	23.2
Neighborhood income quintile
Missing	448	0.9	2,220	0.5	2,356	0.5	5,024	0.5
1 (low)	14,699	28.6	114,695	25.2	96,265	20.0	225,659	22.8
2 (medium-low)	10,529	20.5	94,491	20.7	98,247	20.4	203,267	20.6
3 (medium)	8,494	16.5	86,036	18.9	95,864	19.9	190,394	19.3
4 (medium-high)	8,674	16.9	82,416	18.1	97,320	20.2	188,410	19.1
5 (high)	8,613	16.7	75,621	16.6	90,899	18.9	175,133	17.7
Immigration status
Non-immigrant	46,406	90.2	419,170	92.0	416,526	86.6	882,102	89.3
Immigrant	3,756	7.3	28,738	6.3	54,082	11.2	86,576	8.8
Refugee	1,295	2.5	7,571	1.7	10,343	2.2	19,209	1.9
Rural residence
Missing	85	0.2	358	0.1	379	0.1	822	0.1
No	46,489	90.3	395,333	86.8	416,843	86.7	858,665	86.0
Yes	4,883	9.5	59,788	13.1	63,729	13.3	128,400	13.0
Long-term care residence
No	50,133	97.4	376,160	82.6	461,705	96.0	887,998	89.9
Yes	1,324	2.6	79,319	17.4	19,246	4.0	99,889	10.1
Comorbidity
0 ADGS	1,143	2.2	3,502	0.8	10,887	2.3	15,532	1.6
1–5 ADGS	15,302	29.7	87,677	19.2	152,059	31.6	255,038	25.8
6–9 ADGS	18,153	35.3	149,310	32.8	182,882	38.0	350,345	35.5
10+ ADGS	16,859	32.8	214,990	47.2	135,123	28.1	366,972	37.1

Abbreviation. ADG, Aggregated Diagnostic Group.

Source: Authors’ analysis of administrative healthcare data from the Institute for Clinical Evaluative Sciences.

Note: High-cost patients are those in the 90th percentile of the cost distribution. A Chi-square test for differences between mental health high-cost patients and other high-cost patients for each characteristic showed that all of these differences were significant (p < .0001).

^a Not applicable.

^b Median age (and interquartile ranges): majority mental healthcare, 46 (33–57); some mental healthcare, 70 (55–81); no mental healthcare, 63 (46–76); any healthcare, 66 (49–79).

Patient characteristics

As found in previous work, high-cost patients with majority MHA-related costs were younger (median age of 46 years), were equally distributed by sex, lived in poorer neighborhoods (29% in the low-income quintile), and had high levels of comorbidity (33% with 10+ ADGs) (Table 1). Patients with some MHA-related costs had a median age of 70 years, were mostly female (62%), lived in poorer neighborhoods (25% in the low-income quintile), and had higher levels of comorbidity (47% with 10+ ADGs) (Table 1). This group also included a relatively high proportion of patients living in long-term care homes (i.e. nursing homes), which differed from the other two groups. Compared to the previous group, high-cost patients with no MHA-related costs were slightly younger (median age of 63 years), made up of slightly less females (55%), were equally distributed among income quintiles, and had less comorbidity (28% with 10+ ADGs) (Table 1). All patients lived primarily in urban areas. These findings held for patients in the 95th and 99th percentiles (results available upon request).

Average costs by major cost categories and patient groups

We found large variability in some cost categories across patient groups, namely for long-term care, but less for others, such as ED visits (Figure 1). As found previously, costs for high-cost patients with the majority of MHA-related costs were mainly due to hospitalizations (73%). These patients were mainly hospitalized for psychosis and mood disorders (not shown). For high-cost patients with some MHA-related costs, costs were mostly due to hospitalizations (30%), long-term care (20%), and physician services (14%). MHA-related costs only made up 3.1% of total costs, of which close to half (1.4%) was for outpatient prescription drugs (see Supplementary Figure 1). Patients were mainly hospitalized for COPD, heart failure, renal failure, and dementia (not shown). For patients with no MHA-related costs, costs were mostly due to hospitalizations (37%), physician services (18%), and other care (18%) (in particular, cancer and dialysis clinic visits). Patients were mainly hospitalized for ischemic heart disease (such as acute myocardial infarction) and cancer (not shown). Results changed a bit for the latter two cohorts when we restricted our analysis to patients in the 95th and 99th percentiles (results on the sensitivity analysis for the first group can be found elsewhere⁹). (For the 95th percentile, the proportion of costs with physician care decreased, but increased for long-term and other care for both patient groups; all other categories remained relatively the same. For the 99th percentile, the proportion of costs with hospitalizations and other care increased, but decreased for physician services and drugs for both patient groups; all other categories remained relatively the same [results available upon request]).

Figure 1. Cost distribution across major health services by high-cost patient groups in Ontario in 2012. Abbreviations. MHA, mental health and addiction; ED, emergency department. Physician services include outpatient services, such as lab tests and non-physician services. Prescription drugs include those covered under the provincial public healthcare plan. Other care includes other ambulatory care (same-day surgery, cancer, and dialysis clinic visits), rehabilitation, and complex continuing care. Source: Authors’ analysis of administrative healthcare data from the Institute for Clinical Evaluative Sciences.

Predictors of high spending

Focusing solely on high-cost patients with some or no MHA-related costs, once we controlled for potential confounders, such as socio-demographic characteristics and comorbidity, we found that having any MHA-related costs increased total costs by roughly $2,698 (p-value <.001) (Table 2). This coefficient increased to $3,256 for the 95th percentile, but decreased to –$2,932 for the 99th percentile. Compared to the unadjusted value of $8,181, roughly two-thirds of the cost differential between both patient groups could be explained by observable factors. Other major predictors of costs (in terms of magnitude) included being a male ($3,296; p-value <.001) and living in a long-term care home ($24,285; p-value <.001). This finding held for the 95th and 99th percentiles, but with a negative sign for long-term care for the latter percentile.

Table 2. Conditional marginal effect of socio-demographic and clinical variables on healthcare costs among high-cost patients with some and no mental health and addiction-related costs in Ontario, 2012.

Variable	Conditional marginal effect	Standard error	p-value
Any mental health use	2,697.989	63.400	.000
Male	3,296.600	61.212	.000
Age	668.062	10.322	.000
Age^2	–4.225	0.082	.000
Neighborhood income quintile 2^a	69.505	84.931	.413
Neighborhood income quintile 3	–410.351	84.949	.000
Neighborhood income quintile 4	–653.289	84.322	.000
Neighborhood income quintile 5	–807.823	86.005	.000
Immigrant	–1,002.307	115.834	.000
Refugee	–1,765.644	222.356	.000
Number of comorbidities	539.648	7.909	.000
Long-term care residence	24,284.850	123.319	.000
Urban residence	961.147	80.856	.000
Administrative health region effects	Yes
Number of observations	931,854
Log pseudolikelihood	–8,572,886,896
AIC	18,399.64
BIC	1.71e

^a Neighborhood income quintile 1 is the reference category.

Abbreviations. AIC, Akaike information criterion; BIC, Bayesian information criterion.

Source: Authors’ analysis of administrative healthcare data from the Institute for Clinical Evaluative Sciences.

Next, we examined the impact of particular hospital diagnoses on total costs (Table 3). All hospital diagnosis-related coefficients were statistically significant (p < .001). Diabetes, cancer, renal failure, and stroke were among the chronic conditions with the highest coefficients ($23,833, $20,207, $18,037, and $16,750, respectively). Although a bit lower, having a hospitalization for psychosis was also found to be a major predictor of costs ($15,209). In addition, living in a long-term care home had a large impact on costs ($27,459; p < .001). These results largely held for the 95th percentile with a few changes—the coefficients associated with psychosis and dementia increased, while the coefficient with substance abuse became negative. For the 99th percentile, having a hospitalization for psychosis, diabetes, stroke, renal failure, dementia, or heart failure increased costs, while having one for substance abuse, cancer, mood disorders, ischemic heart diseases, or COPD decreased costs (effects are listed from largest to smallest).

Table 3. Conditional marginal effect of socio-demographic, clinical, and diagnostic variables on healthcare costs among high-cost patients with some and no mental health and addiction-related costs in Ontario, 2012.

Variable	Conditional marginal effect	Standard error	p-value
Male	2,116.813	57.833	.000
Age	512.203	9.878	.000
Age^2	–3.198	0.078	.000
Neighborhood income quintile 2^a	–35.320	80.235	.660
Neighborhood income quintile 3	–467.887	80.210	.000
Neighborhood income quintile 4	–693.501	79.537	.000
Neighborhood income quintile 5	–809.936	81.325	.000
Immigrant	–1,122.890	109.182	.000
Refugee	–1,471.310	210.378	.000
Number of comorbidities	553.277	7.222	.000
Schizophrenia and other psychoses	15,209.190	2,002.286	.000
Mood (affective) disorders	13,264.940	1,073.027	.000
Dementia	9,282.129	521.757	.000
Substance abuse	4,586.978	720.103	.000
Cancer	20,207.370	182.228	.000
Diabetes	23,832.550	520.057	.000
Ischemic heart diseases	10,021.990	161.098	.000
Heart failure	13,228.310	256.204	.000
Stroke	16,749.770	342.131	.000
COPD	7,616.076	206.107	.000
Renal failure	18,036.700	454.274	.000
Presence of both mental and physical conditions^b	10,012.390	604.712	.000
Long-term care residence	27,459.420	123.517	.000
Urban residence	1,012.789	76.668	.000
Administrative health region effects	Yes
Number of observations	931,854
Log pseudolikelihood	–7,789,179,197
AIC	16,717.60
BIC	1.56e

^a Neighborhood income quintile 1 is the reference category.

^b Based on hospitalizations with diagnoses for mental and physical health conditions.

Abbreviations. COPD, chronic obstructive pulmonary disease; AIC, Akaike information criterion; BIC, Bayesian information criterion.

Source: Authors’ analysis of administrative healthcare data from the Institute for Clinical Evaluative Sciences.

All results were largely unchanged, with the exclusion of drug costs covered under the public provincial drug plan for patients younger than 65 years.

Discussion

Managing the care of high-cost populations is a key concern for policy-makers. Our work demonstrates that there is considerable heterogeneity among this population, namely in terms of their socio-demographic characteristics, chronic conditions, and, thus, how they utilize the healthcare system. Previous work found that high-cost patients with majority MHA-related costs incur over 30% high costs than other high-cost patients⁹. In this analysis, we found that patients with some MHA-related costs incur higher mean costs compared to high-cost patients with no MHA-related costs, even after controlling for relevant socio-demographic and clinical variables. Furthermore, we found that having a hospitalization for psychosis increases costs substantially and is comparable, from a cost perspective, to having a hospitalization for stroke, thus highlighting the non-trivial role of mental disorders among high-cost patients without high MHA-related costs.

Our results are in accordance with existing research. One study, which examined all Medicaid high-cost patients in Maryland, found that serious mental illness was among the most frequent diagnoses for all high-cost patients with hospitalizations²⁷. Other work examining mandatory Medicaid managed care patients in New York also found that the level of mental illness among this population was quite high¹⁰.

The disproportionate use of health services by a small proportion of patients is inevitable and likely appropriate in a healthcare system that provides need-based care. Nonetheless, the Commission on the Reform of Ontario’s Public Services has recommended that efficiencies be found in healthcare for these patients²⁸. Prior work has shown that policies and interventions designed to address quality of care and high healthcare spending in general will likely not apply to high-cost patient sub-groups^6,9. We found that a substantial group of high-cost patients had some type of MHA-related care. Furthermore, we provide evidence that many high-cost patients have relatively high levels of comorbidity, both physical and mental, making them a complex patient population. There is an increasing need to better understand complexity and comorbidity, including psychiatric comorbidity, and a recognition that current, single disorder approaches to healthcare delivery fall short on guaranteeing good outcomes among individuals with multi-morbidity²⁹. This recognition has become a priority in Ontario, Canada, as evidenced by the establishment of the Medical Psychiatry Alliance in 2014. The goal of this collaborative partnership is to transform the delivery of mental health services for patients who suffer from physical and psychiatric illness or medically unexplained symptoms³⁰.

High-cost patients with majority MHA-related costs are younger (median age of 46 years), often diagnosed with psychosis or mood disorders, and their costs are mainly due to psychiatric hospitalizations. Previous research has provided recommendations on how best to manage these patients, which include regular post-discharge physician follow-up and medication adherence⁹. High-cost patients with no MHA-related costs had a median age of 63 years, circulatory and/or digestive systems-related diseases, high levels of comorbidity, and costs mainly due to hospitalizations and other care (such as visits to cancer care and dialysis clinics). This patient group requires interventions that include disease management and care co-ordination⁶. High-cost patients with some MHA-related costs were older, with a median age of 70 years, mostly female, with some mental disorders and/or addictions, circulatory and/or digestive systems-related diseases, very high levels of comorbidity, and costs mainly due to hospitalizations and long-term care. This patient group requires strategies that improve timely access to MHA care, and the integration of MHA services into broader care co-ordination and disease-management models⁶. Across patient sub-groups, physicians have a key role in engaging high-cost patients and matching them with specific interventions or programs according to their clinical profiles and needs⁶.

Our study made use of rich administrative healthcare data and included a comprehensive population-based sample of all adult high-cost patients in Ontario, Canada’s largest province. Few studies have been able to examine characteristics and costs among all adult high-cost patients within a given jurisdiction. We accounted for the vast majority of healthcare costs covered by the Ontario ministry of health (roughly 91%) under a comprehensive universal healthcare insurance plan³¹; most studies only examine costs of some health services. Despite differences in the delivery and funding of healthcare, these findings may be applicable to other jurisdictions in the developed world³².

A few limitations are worth noting. We did not examine or model costs for patients with a majority of MHA-related costs, as these were covered in other work⁹. We did not examine children and adolescents, as some services for this population are funded by ministries other than the Ontario Ministry of Health and Long-term Care. We were not able to account for addiction-related healthcare costs from community-based agencies, where a large part of addiction treatment is provided³³, as these data were not available. This likely biased our estimates of publicly-funded healthcare costs downwards. We were able to capture outpatient drug costs for patients covered under the public provincial drug plan only. These last two limitations may have contributed to instances where patients were misclassified as not having MHA-related costs due to the lack of data on addiction treatment and/or on drugs covered under private healthcare plans. Finally, our analysis made use of only one year of data, which were the only data available at the time of analysis. Previous work has found that some high-cost patients do not persist in that category³⁰. Thus, future research will make use of longitudinal data to understand which patients are more likely to remain in the high-cost category over time and whether trajectories of healthcare use and cost differ by patient group.

Conclusion

This analysis provides relevant information on high-cost populations. High-cost patients are a heterogeneous group and should not be considered as a whole when designing potential interventions/policies. To date, few studies have examined this issue using a comprehensive population-based sample. We found that high-cost patients with some MHA-related costs had higher average costs than those without, different patient characteristics, chronic conditions, and healthcare use. In particular, patients with some MHA-related costs were older (median age of 70 years) and mostly female, had circulatory and/or digestive systems-related diseases and some mental disorders and/or addictions, and costs mainly due to hospitalizations and long-term care. On the other hand, patients with no MHA-related costs were slightly younger (median age of 63), had circulatory and/or digestive systems-related diseases but no mental disorders and/or addictions, and costs mainly due to hospitalizations and other care. Both patients were in contrast to high-patient with majority MHA-related costs, who were substantially younger (median of 46 years), had mostly mental disorders, and whose costs were mainly due to (psychiatric) hospitalizations. More research is required to understand the implications of heterogeneity among high-cost patients in general, and of MHA-related comorbidity more specifically. However, these findings suggest that interventions/policies targeted at high-cost patients should try to address both their physical and mental health, where appropriate.

Transparency

Declaration of financial/other relationships

The opinions, results, and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by the ICES or the Ontario MOHLTC is intended or should be inferred. Parts of this material are based on data and information compiled and provided by the Canadian Institute for Health Information (CIHI). However, the analyses, conclusions, opinions, and statements expressed herein are those of the authors, and not necessarily those of CIHI. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgment

This study was supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (MOHLTC). This research was supported in part by the Medical Psychiatry Alliance (MPA), a collaborative health partnership of the Centre for Addiction and Mental Health, the Hospital for Sick Children, Trillium Health Partners, the University of Toronto, the Ontario Ministry of Health and Long-Term Care and an anonymous donor. The MPA's goal is to provide better access to, and coordination of, integrated health care services for patients living with co-existing mental and physical illnesses.