This study is the first to estimate the lifetime costs of stroke in Korea. Previous studies mostly measured the annual cost of stroke using a prevalence-based approach or were limited to the estimations of costs over the short-term (one or several years following the event) [27, 33]. Since half of the survivors from stroke are left with permanent disability and consequently bear a lifetime burden, an incidence-based approach has the advantage of providing a more comprehensive picture of the health and economic impact of stroke .
The present study showed how the cost of stroke varied according to the age of onset. As commonly observed in earlier studies, per-capita lifetime costs were greater in younger stroke patients mainly due to long-term management and loss-of-productivity costs . In the present study, the expected lifetime cost of stroke for men experiencing acute stroke at the age of 45 years was estimated as 200.7 million KRW, whereas that for men 65 years of age was 16.4 million KRW, a difference greater than 12-fold. Although the difference was less compared to that of men, the lifetime costs for women suffering stroke at 45 years (75.7 million KRW) were still approximately 3.9 times higher than those for women suffering stroke at 65 years (19.3 million KRW).
It has been observed that the average age of new-onset stroke is decreasing in Korea. While the proportion of stroke patients aged in their 70s and 80s has increased by 22.84% and decreased by 4.86%, respectively, between 1998 and 2005, the proportion of patients in their 40s and 50s has increased by 90.91% and 54.76%, respectively . The trend toward younger stroke patients in Korea raises a concern for the national economy because of an erosion of the work force. Additionally, the present study showed that 75% of the national lifetime costs of stroke were attributable to patients aged less than 65 years, although this age-group accounted for only 30% of incident strokes (Figure 2). Consequently, if the average age of stroke patients in Korea continues to decline, then the economic burden of stroke is expected to increase substantially.
Lifetime costs of stroke are greater among men than women due to greater indirect costs (i.e., loss of productivity). The average wage rate in Korea is higher for men than women. Another possible explanation is that the value of women working as housewives and caregivers was not considered, resulting in underestimation of the productivity loss in women. Notably, medical costs were also higher among men than women throughout the different age groups, which is different from data from other countries. For example, according to the summary of published international data between 1993 and 2003 on the costs of stroke, indirect costs were greater for men but direct costs were greater for women . Because it is unknown whether males suffer more severe strokes than females, an unbalanced utilization of healthcare between the genders is possible.
Two methods are available to estimate health-related productivity changes. The human capital approach, which is widely used, is the traditional approach to measure and valuate the potential loss of productivity for all income lost due to absence from work and premature death. This approach is criticized because the true cost imposed to society is usually overestimated . The friction cost approach assumes workers will be replaced after a certain period of time, called the friction period, and the amount of lost productivity depends on the time and cost of replacement. This approach attempts to measure the actual loss of productivity in the economy and usually provides much lower estimates for the loss of productivity than do the estimates from the human capital approach [28, 30]. However, we prefer the human capital approach for several reasons. First, the objective of the present study was to evaluate the value of a lost and damaged human life due to stroke rather than estimating the actual monetary loss of productivity in the industry. Second, economic activity returns back to the natural rate of unemployment in the long-run according to the macroeconomic theory. Lastly, there are too many uncertainties to estimate the actual loss of productivity using the friction cost approach because the data about the friction period and employer costs (i.e., hiring and training costs) are difficult to estimate. To obtain the estimates of employer costs needed for the friction cost approach, we searched published literatures and reports intensively to find Korean data, but failed to identify the needed estimates. Government organizations in Korea were also contacted (departments related to human resources and vocational training in the Ministry of Strategy and Finance and Ministry of Employment and Labor) to seek expert opinions on the estimates; however, these searches were unsuccessful. Fortunately, we were able to find the published report of the American Society for Training and Development which contained international comparisons regarding employee training. However, since the report did not include data for Korea only, the estimates for Asia (an average of 362 US dollars per employee on training in 2000) were used, which included nine countries where the number of included Korean companies responding to the survey was not known . As expected, the friction cost approach gave us produced much lower estimates for the loss of productivity (23% and 41% of the expected lifetime costs of stroke estimated by a human capital approach for men experiencing acute stroke at the age of 45 and 55 years, respectively).
One of the major strengths of the present study is that the estimation of the lifetime cost of stroke was based on national data, both in terms of unit costs (i.e., claims records of NHI) and epidemiologic information. This provides greater generalizability of results, compared to that of earlier studies based on patient records from selected hospitals or local stroke registries [10, 24]. Another strength of the study was the application of age- and gender-specific cost and epidemiologic data. Survival rate, recurrence rate, and cost of stroke treatment are well known to be affected by patient age and gender .
Several assumptions were made in the Markov model that may have led to over or underestimation of stroke costs. First, due to the lack of data, it was assumed that the cost for a recurrent stroke was the same as the cost for the first stroke. However, if the cost of treating a recurrent stroke was more expensive than the first-ever stroke due to the increased severity, then the model would have underestimated the true cost. Secondly, in estimating the costs associated with premature death, we assumed that there was no loss of productivity costs after the retirement age of 65 years. This conservative assumption may also have contributed to the underestimation of the true costs of stroke. On the other hand, the universal application of transportation and loss of productivity costs to all hospital admissions and office visits due to stroke may have led to the overestimation of the true costs of stroke. Additionally, the choice of a human capital approach instead of a friction cost approach in estimating loss of productivity cost would cause overestimation of the costs.
The present study has several limitations. First, although the study has high external validity because of the use of NHI claims data from the entire population in Korea, case ascertainment for stroke solely based on diagnostic codes of insurance claims records may be less accurate compared to a community-based incidence cohort study [10, 24]. The validity issue of administrative database coding for stroke cannot be easily resolved since the database is constructed for reimbursement purposes and not for clinical purposes. However, the NHI claims data has a higher reliability than any other data in Korea to estimate the healthcare utilization and costs in terms of representativeness and generalizability since the NHI claims data include the entire Korean population. Moreover, the accuracy of diagnosis codes tended to be higher for claims of severe conditions such as stroke compared with those of other mild conditions [35, 36]. In addition, all claims data in Korea is submitted electronically, thus the precision of data is higher than that of the paper-based claims data. Second, due to the lack of detailed epidemiologic data, we applied the same transition probability of having a recurrent stroke regardless of how many cycles or years passed since the initial stroke. Third, information on patients' out-of-pocket expense outside the hospital may have limited generalizability since it was derived from patient survey results from a selected hospital. Fourth, we were not able to incorporate the distribution of severity of stroke measured by neurological scales and functional outcome scales into the analysis because these clinical values were not available in the claims database. Ignoring the distribution of stroke severity may cause distortions in the estimation results of stroke impact.
Last, we did not reflect the value of productivity loss of housewives with stroke. Moreover, we did not consider the productivity loss of family caregivers, who are usually women who do not work outside the home and take care of the stroke victims that are unable to work and do not stay in hospital. Accordingly, the value of stroke in Korea may have been underestimated.
Further research is needed to examine whether the true costs of treatment for stroke patients are different between the results obtained from the claims database and from the patient registry data which include clinical variables representing the severity of stroke.