The one-year healthcare costs attributable to ischemic stroke was examined by calculating the marginal estimated healthcare Medicare payment difference between a 2004 ischemic stroke cohort and a propensity matched 2004 non-stroke control group. The estimated marginal 2004 cost of stroke-related healthcare in the year after ischemic stroke in SC was $9,054. This amount reflects the difference between the annualized healthcare cost of the stroke group of $27,330 less the healthcare costs of the propensity score matched non-stroke control group of $18,276 (p-value <0.0001). Similarly, the 2004 marginal cost of stroke-related rehabilitation services in the year after ischemic stroke was $2,651, resulting from a statistically significant difference between the average annualized total rehabilitation services cost for the stroke group of $3,735 and the control group of $1,064 (p-value <0.0001). The findings clearly demonstrated that the total cost of healthcare after ischemic stroke significantly overestimates the cost of stroke as an illness because the marginal cost of stroke in this older population is attenuated by healthcare costs related to comorbidities.
While these amounts are substantial and statistically significant, they are lower than estimates commonly reported. The most frequently cited estimates are by Taylor, Davis and colleagues (cited in 559 publications referencing this research (accessed May 2012)) . Interestingly, the Taylor study continues to be referenced in the “Heart Disease and Stroke Statistics—annual update: A report from the American Heart Association” [14–18], even though this research is based on 1990 data and does not benefit from recent improvements to methodology. This continued referencing is likely the result of no new estimates emerging in the literature using population-based studies on the cost of stroke.
Further, the authors of the Taylor study estimated the 1990 annual direct cost of ischemic stroke in the 65–74 year age group to be $17,823 ($35,197 in 2004 dollars) versus the annual healthcare expenditures for their control group of $2,825 ($5,579 in 2004 dollars) . The Taylor study reports the annual direct marginal cost of ischemic stroke of $14,998 in 1990 dollars ($29,618 in 2004 inflated dollars) which is much higher than we found in the current research. Differences between the current research and that undertaken by Taylor and colleagues in their 1997 publication are likely related to differences in the methods used to estimate these dollar amounts and constantly changing healthcare practice patterns.
A significant trend toward shorter hospital average length of stay has been frequently reported in the literature which is also indicative of changing practice patterns and third-party payer reimbursement changes over time . For example, the Taylor study reports that 70% of the medical costs for the first year after a stroke can be accounted for by the initial hospitalization ; however, the article does not report the average length of stay. In the current study, initial hospital costs only account for 35% of the first year of total healthcare costs. This may reflect the declining length of hospital stays [19–21]. Shorter length of stay would result in lower first year healthcare cost estimates in later studies, however these reductions may be counteracted by inflation adjustment over time.
The Taylor et. al. study includes a 5% Medicare sample of the 1990 US population, while the current study uses all SC Medicare patients with ischemic strokes in 2004. Taylor et. al. also used an average cost to charge ratio for all admissions to estimate stroke costs. Estimates based on cost to charge ratios for all admissions can skew costs because stroke costs may not follow general hospital cost to charge trends. The current research uses actual payments made by Medicare to the provider and does not make any assumptions in the costing methods. Also, the Taylor research used a 1 in 1,000 randomly selected, non-matched control group from the general US Medicare population. This practice would likely under estimate the control groups’ healthcare expenditures because the average Medicare population is more likely to be healthier than a control group matched on an equally at-risk population. This type of matching approach would result in inflated marginal cost estimates. The current study addresses the potential selection bias caused by unmatched controls by using propensity score matching to match the stroke cases to controls, which results in a more conservative and reasonable marginal cost estimate. An added benefit of the matching approach used in the current study was a reduction in the selection bias of other unmeasured factors that are correlated with the known covariates used in this matching algorithm. This gives confidence that the observational study is well matched and is unlikely to contain much selection bias.
In a seminal study by Samsa and colleagues, the 2-year cost and survival after cerebral infarction was estimated based on 1991 data . This publication, which has been cited 130 times (accessed May 2012), reports the first year cost of first ever ischemic stroke in the over 65 Medicare population as $29,444 (in 1991 dollars) . The Samsa estimate is very similar to the current study’s 1-year cost estimate of $27,330, but the current study is based on 2004 Medicare payments while theirs is calculated using 1991 data. The Samsa study estimated total average cost rather than marginal cost. In 1991 the average length of hospital stay after acute stroke was considerably longer than it was in 2004 , resulting in inflated estimates when compared to current practice. Their estimates also used cost to charge ratio adjusted Medicare charges for all facility bills, rather than payments, which has an unknown effect on the costs incurred by Medicare. Thus, the estimates provided by Samsa et. al. are no longer applicable to the current healthcare system and do not take into account the cost of the non-stroke care that surviving stroke patients are likely to incur even if they had not had a stroke. However, the Samsa study provides valuable insight to the inflated estimation of costs that result from reporting total healthcare cost instead of marginal cost of care, particularly in the group examined in the current study. That is, older individuals tend to have significant healthcare costs outside of the particular major illness being studied, which make it essential to take into account other costs when estimating the costs attributable to a single disease.
There are many policy implications related to the use of inflated estimates for stroke. It is generally believed that commonly quoted figures related to incidence rates, prevalence rates, proportional estimates of certain services and cost estimates are inflated. However, the public, including those in the research community, continue to use and report those same numbers. This practice may negatively affect the allocation of funding for other important research since there are limited dollars to support health-related research efforts. Furthermore, when advocacy and governmental groups quote these inflated estimates they lose credibility with the public and their ability to make good policy decisions are diminished. Also, given the climate of cost-containment in health care, the accuracy of estimates of total cost of care related to any particular condition can help providers and funding groups better plan for resources to provide such care.
There are several limitations in the analysis of SC Medicare claims data. There is an inherent limitation of using administrative claims to ascertain diagnoses and identify resource use and costs with complete accuracy, because these data are not purposely collected for clinical research but are collected for the distinct purpose of making healthcare payments. It is possible that signs or symptoms related to stroke may not have been captured in the claims, and that ischemic stroke patients not diagnosed with 434.xx and 436.xx ICD-9-CM codes under the primary diagnosis category would be missed by this analysis. Similarly, the measurement of rehabilitation and general resource use in these data depends on the design and implementation of the Medicare fee-for-service plan and its scope of coverage which may change over time. Medicare reimbursement rules and healthcare practice patterns that change over time may also make these 2004 based results less generalizable to current times. The use of SC estimates from this research may not be representative of stroke costs and rehabilitation utilization in other states and in the US since healthcare practice patterns and costs vary geographically. In addition, the interpretation of rehabilitation resource use and the assignment of associated costs are challenging.
Due to the lack of availability of clinical information in these claims data, the classification of rehabilitation was based on the inclusion of specific rehabilitation related codes in the ten diagnosis code and four procedure coding columns provided in the data files. Provider billing systems will often allow for a much longer list of these codes when interpreting medical records into billing data in the clinical setting. So it is reasonable to assume that some diagnoses or procedures related to stroke and rehabilitation would not be included in the Medicare data.