ESTIMATING DISEASE SPECIFIC COST OF AN INSURED POPULATION USING HEALTH INSURANCE CLAIMS
[paper presented at the 2nd UK-Japan epidemiology seminar held in Osaka, August 1996]
AtoZ OKAMOTO (Department of Public Health, Kinki University Medical School)
INTRODUCTION
Limitations of traditional classification method
Breaking down the health care cost of an insured population into disease specific cost provides fundamental information for analyzing how the money is spent for health care services and developing health policy in both local and national levels.
For patients with a single diagnosis, the cost spent for a given patient can safely be assumed that it was spent entirely for the specific disease. However, for patients with multiple diagnoses, the task of breaking down the entire cost into subcategories becomes more complicated and difficult.
Ideally patients' medical records should be obtained and reviewed item by item to classify the cost, for example, pharmaceuticals and examinations. Obviously this procedure is time-consuming, labor-intensive and requires professional skills to warrant reliable outcome. Moreover, for all-inclusive services such as physicians' medical management or nursing, the attempt to break down into different disease category itself is impossible.
Traditionally, epidemiological survey using health insurance claims has employed "primary diagnosis" approach, in which claims with multiple diagnoses will be assigned a single "primary diagnosis" and be classified accordingly. Social Insurance Claims Survey (SICS) conducted by Ministry of Health & Welfare (MHW) every year defines "primary diagnosis" as "a diagnosis considered to have consumed the largest share of the entire cost".
This approach is simple and requires less energy for data inputting. However it has some limitations: Once a claim is classified into one of the multiple diagnoses, the entire cost spent for a given patient is considered to have been spent exclusively for the selected diagnosis, while actually it is not. Also the procedure to determine a primary diagnosis out of multiple diagnoses requires professional skills and therefore time-consuming. And even though the procedure is conducted by well-trained reviewers, it can not be free from individual variances.
PDM method
In order to overcome the above limitations of traditional "primary diagnosis" approach, the author proposed Proportional Disease Magnitude (PDM) method, in which the entire cost of a claim will be objectively broken down in proportion to the "magnitude" preassigned to each diagnosis[1].
The preassigned magnitude used in PDM method is calculated in each disease category by multiplying (1) the likelihood of the diagnosis to be primary diagnosis and (2) the cost per claim classified as the diagnosis.
(1) is obtained from Patient Survey, a nationwide sampling survey conducted by MHW every three years, in which doctors of sampled hospitals or clinics respond the primary and secondary diagnosis of patients. The likelihood of a diagnosis to be primary diagnosis is expressed as (a)/((a)+(b))
whereas (a) denotes the number of patients with the diagnosis as primary diagnosis (this includes both patients with single diagnosis and patients who have multiple diagnosis), and (b) denotes the number of patients with the diagnosis as secondary to the other primary diagnosis.
(2) is obtained from SICS, a two-tier stratified sampling survey on health insurance claims. The author admits committing a contradiction by using data developed by traditional "primary diagnosis" method as source for new PDM method. This may be excused by the fact that there has been no survey data of reliable sample size utilizing PDM method, and that the sample size of SICS is large enough (30,693 inpatient claims and 265,070 outpatient claims for May 1993)to warrant its use as source data for the new method.
PDM method will provide an objective and consistent method to break down the cost of insurance claims thereby yielding disease specific cost estimate for every disease category. It can also provide disease specific service volume for such as inpatient days or the number of examinations. However it has not been field-tested to verify the accuracy and reliability of the product.
In this article, the author attempted to field-test the PDM method and evaluated the product against traditional method.
METHODS
[study design] crossover study on the same sample to compare traditional and new methods.
[data used] type: inpatient insurance claims (submitted by providers on services rendered in every calendar month) source: National Health Insurance run by A city (retirees segment)
patient characteristics; retirees and their dependent family members aged 60 through 69 at the time of service.
time of service: May 1986
sampling method: population, sample size: 99
[data extraction] For the purpose of disease classification, B class category of ICD-9 (commonly known as "99 classification" for epidemiological survey using health insurance claims) was used. All diagnoses appearing in the "list of diagnoses" in claims were coded with two digits number(B1-B99). Provisional diagnoses designated as, for example, "suspected of" were intentionally excluded from this study because most of such diagnoses are assigned without firm evidence just to avoid "medically unnecessary denial of payment". The total cost expressed in the unit of 10 yen was also extracted from the sample claims.
[PDM method] Disease specific "magnitude" was calculated for each B class category using cost data of inpatient claims from SICS conducted in June 1986 (surveyed claims were predominantly from previous month) and data on primary vs. secondary diagnosis from Patient Survey conducted in October 1987. The data from Patient Survey comprised both inpatient and outpatient, which will limit generalizability of this study findings (Patient Survey conducted in 1990 and 1993 provides primary vs. secondary diagnosis data broken down by in- and outpatient, however it is given in an aggregated category because of larger standard error). The total cost of each claim was divided in proportion to the magnitude preassigned to each diagnosis, and the cost for each diagnosis was summed up to yield an estimate for disease specific cost of the sample claims. [traditional method] For the sake of comparison, cost estimate based on traditional method was conducted by three qualified physicians. Each physician reviewed the sample claims independently and chose the primary diagnosis. The cost for a claim was considered to have been spent exclusively for a primary diagnosis chosen by reviewers.
RESULTS
There were 518 diagnoses contained in 99 claims with a mean of 5.22 diagnoses per claim (SD=2.99). Six claims had single diagnoses and hence posed no classification problems. The maximum number of diagnoses was 15. The total cost spent for the sample claims was 36,840,350 yen.
The result of traditional classification method showed that three independently reviewing physicians agreed on the B class category of primary diagnosis in 43 claims out of 99 (it should be noted that even if physicians chose different diagnoses, it is still treated as "agreed" if the chosen diagnoses fall in the same B class category). Two of three physicians agreed in 42 claims and all of them disagreed in 14 claims.
Disagreement increased as the number of diagnoses in a claim increases: the claims for which three physicians agreed had an average of 4 diagnoses (s.e.=0.415) while the claims for which two physicians agreed had an average of 5.738 diagnoses (s.e.=0.463) and those for which all three disagreed had an average of 7.5 diagnoses (s.e.=0.492).
The estimated cost for all category is expressed as percent of the total cost in [Table.1].
DISCUSSION
As expected, disease classification by PDM method is more distributed over entire categories than traditional method.
Diseases that are less likely to be primary diagnosis tend to be ignored when classified by traditional method. For example, nutritional and endocrine disorders including hyperlipidemia (B20)and anemia (B21) are mostly disregarded by traditional method while PDM method estimates that more than 5% of total cost was spent for these ailment.
On the other hand, more serious diseases such as cancer tend to be underestimated by PDM method. This reflects a difference of ideology between two methods: the traditional primary diagnosis approach assumes that all services are rendered in relation to a single primary diagnosis while PDM method try to separate the services based on the indication of each services. For example, if a patient with stomach cancer develops anemia and is checked CBC count and given medicine for anemia, traditional method assumes that all those services were rendered for cancer treatment although CBC and anemia medicine per se are not indicated for cancer. PDM method is interested in estimating how much service and cost were utilized for specific diseases or conditions.
PDM method will enable insurers to get a quick and easy estimate of how cost and services were utilized for each disease category. With full computerization of insurance claims slated in the near future, PDM method presents strengths over traditional method in that it can be performed without professional skills, time and any biases inherent to reviewers individual variances.
[Table.1] Break down of the health care cost into disease category
N=99 claims, total cost=36,840,350 yen
REFERENCES
[1].AtoZ Okamoto. Proportional Disease Magnitude (PDM) Method for Computerized Health Insurance Claims. Kosei-no-Shihyo, July '96 issue [in Japanese, in press].
BACK
BACK to TOP
GO to CareManager.Net