The RAND Health Insurance Experiment was the first scientifically rigorous study that established the impact of cost-sharing for medical services.¹ Interim results showed that the total cost of health care for participants who received free care was 38% higher than that for families who paid for part of their care.² On completion of the RAND study, there was strong evidence to suggest that ill patients were less likely to seek treatment or be admitted to hospital if they had to pay a portion of the cost.^1,3,4

The most comprehensive, recent study to examine the impact of cost-sharing specifically of prescription drugs investigated various employer drug-benefit plans and prescription drug spending.⁵ Beneficiaries of the plans with higher co-payments used less medication and less expensive drugs than those with lower co-payments, and although overall costs for insurers fell, the beneficiaries' out-of-pocket drug expenses rose significantly. In a managed-care setting, studies have shown that cost-sharing was associated with reduced expenditures for prescription drugs and for physician services and outpatient hospital services.^6,7,8 The results of these studies suggested that prescription drugs and other health services were “complementary” goods. (Goods or services are considered to be “complementary” when the increase or decrease in the price of one of the goods or services leads to a corresponding decrease or increase in the consumption of the other. In contrast, goods and services are defined as economic substitutes if the quantity demanded of one rises or falls when the price of the other rises or falls.)

In Canada, where most treatments and health services are covered by the government and user fees are prohibited, a predominantly publicly funded health care system with first-dollar coverage (no deductible or co-payment) exists. One gaping hole in the Canadian health care coverage blanket is that outpatient drugs are not available free of charge to all people. At present, outpatient drug coverage varies considerably; for example, people on social assistance usually do not have to pay for their prescription drugs. The overall impact of having incomplete coverage for drugs remains unknown.⁹

BC PharmaCare is British Columbia's drug benefit plan that assists provincial residents in paying for eligible prescription drugs and designated medical supplies. The plan provides residents different levels of coverage. At the time of the study, BC PharmaCare required that people aged 65 years and older pay 100% of the dispensing fees, but not the ingredient costs, to an annual maximum of $200. Once this deductible was reached, the drug benefit plan covered all dispensing fees and ingredient costs of drugs on the provincial formulary for the remainder of the year. PharmaCare has a Maximum Days Supply Policy to prevent wastage and stock-piling. For short-term or first-time drug prescriptions of maintenance drugs, the coverage is limited to a maximum 30-day supply, whereas repeat prescriptions of maintenance drugs are covered for a maximum 100-day supply.

For this study we selected elderly patients from the population-based cohort described earlier who had existing rheumatoid arthritis (diagnosed in the previous year or earlier) and who had reached the annual maximum co-payment of $200 for any calendar year from 1997 to 2000. For each patient selected, we included in the analysis only data for health care services used during the year(s) in which the annual maximum co-payment had been reached. Since the sample was selected on the basis of annual drug prescription records and the year rheumatoid arthritis was first diagnosed, the number of patients being selected differed from year to year.

The outcomes assessed were whether patients were admitted to hospital, the number of hospital admissions, the number of physician visits and the total number of prescriptions filled. All of these outcomes were measured for all care received.

We assessed the following variables for rheumatoid arthritis care as covariates: whether patients had been prescribed disease-modifying antirheumatic drugs (DMARDs), an important part of the medical treatment of rheumatoid arthritis; how long patients had the condition; whether patients were prescribed oral corticosteroid therapy during the study period; whether and how often patients visited an orthopedic surgeon; whether they had undergone orthopedic surgery; and the number of orthopedic procedures performed for rheumatoid arthritis.

For statistical analysis, we defined the period during which the annual maximum co-payment had not been reached as the “cost-sharing period” and the one beyond which the annual maximum had been reached as the “free period.” For each of the calendar years from 1997 to 2000, we compared the patients' utilization of all health care services between the cost-sharing and free periods.

For our analysis of the use of hospital services, we used a 2-part model. First, we compared the probability of a patient being admitted to hospital during the cost-sharing period with that of admission during the free period. For this analysis we used an estimated logistic regression analysis folowing the generalized estimating equations (GEE) approach (with an unstructured correlation matrix).¹² Second, we studied the number of hospital admissions in a given year among the patients who were admitted during the year. To adjust for the different lengths of the cost-sharing and free periods, we used monthly figures. A paired sample comparison was performed for each year to assess the difference in monthly utilization of hospital services. The 95% confidence interval for each of the mean differences was estimated through bootstrap sampling. We also fitted a mixed-effect model with a random patient effect and fixed effects of period and year and other explanatory variables (e.g., age, sex, DMARD use, corticosteroid use, visits with an orthopedic surgeon and performance of orthopedic surgery). The covariate values were evaluated based on data since the beginning of 1996 to the end of the year before which the outcomes were evaluated. Utilization in a particular year, the dependent variable, was modelled to be dependent on covariates from the previous years except for the dummy variable indicating the cost-sharing period and the free period; therefore, utilization was based on data from 1997 to 2000, but covariates were evaluated using data from 1996 to 2000. Explanatory variables were retained in the final model if p values for the β coefficients were less than 0.05.

The annual utilization of hospital services, physician services and prescription drugs are summarized in Table 2. About 54% of the patients were admitted to hospital at least once each year. Patients who were admitted had on average 2 hospital admissions per year. The mean number of physician visits was about 32 per year, and the mean number of prescriptions filled ranged from about 49 to 56. The average dispensing fee per prescription was $7.50, and it took a patient about 81/2 months to reach the annual maximum co-payment of $200.

Table 3 shows the utilization rates during the cost-sharing and free periods. Overall, the proportion of patients with at least one hospital admission during the cost-sharing period was consistently higher than the proportion during the free period over the 4 years. However, the cost-sharing period was about twice as long as the free period (8.5 v. 3.5 months). The number of admissions per month was higher during the cost-sharing period than during the free period except in 2000; the differences were not statistically significant except in 1997. When we compared the number of physician visits per month during the 2 periods, we found that patients used significantly more physician services during the cost-sharing period than during the free period over the 4 years. When looking at the number of prescriptions filled per month, we found that fewer were filled during the cost-sharing period than during the free period. Similarly, patients had fewer prescriptions filled per physician visit during the cost-sharing period.

The results from the multivariate analysis are shown in Table 4. The estimated logistic regression model revealed that the probability of being admitted to hospital did not differ between the cost-sharing and free periods after adjustment for the duration of each period and for other covariates, including age, sex, corticosteroid use, visit to an orthopedic surgeon and year. Among the patients who were admitted to hospital, the adjusted mean difference between the cost-sharing period and the free period in the number of admissions per month was small but was statistically significant. The adjusted mean differences between the 2 periods in the number of physician visits and the number of prescriptions filled were statistically significant, with more patients seeing physicians per month and fewer having prescriptions filled per month and per physician visit during the cost-sharing period than during the free period.

Our findings suggest that prescription drugs and other health care services are economic substitutes. Other recent studies both contradict and support our findings.^6,13 Although Escarce and colleagues⁶ found that prescription drug cost-sharing in US settings was associated with a reduction in expenditures for prescription drugs, as was observed in our study, they found that cost-sharing was associated with reductions in physician services and hospital services as well. They therefore concluded that prescription drugs and these services were economic complements. Tamblyn and colleagues,¹³ on the other hand, observed that, after the introduction in 1996 of a co-payment policy for prescription drugs for elderly people in Quebec, there was an increase in the frequency of hospital, emergency department and doctor visits, along with a reduction in the use of prescription drugs. Given that the latter study was based in Canada, their results and ours are probably more representative of the Canadian setting.

Some potential limitations of our study should be noted. The one limitation shared by all studies using administrative data for the identification of cases is the uncertainty around the accuracy of the diagnosis.¹⁰ Another potential limitation is the presence of supplemental insurance. However, since patients with supplemental insurance would have had it in both periods, we did not expect this factor to be a confounder. Similarly, patients may have used alternative medicines or alternative health care services as a substitute to prescription drugs, which we would have been unable to capture from the administrative data. In addition, we did not have direct measures of disease severity; instead, we used duration of rheumatoid arthritis, use of DMARDs, use of oral corticosteroid therapy, visits to an orthopedic surgeon and orthopedic surgeries performed for rheumatoid arthritis as indicators of disease severity.