Announcement: SOA releases October 2019 Exam STAM passing candidate numbers.

Health Care Variability

Health Care Variability

by John P. Cookson

When adjusting health care costs, variability is a factor that must not be forgotten.

Health Actuaries have been aware of, and have adjusted for, variation in health costs (in dollars spent per capita) by geographic area, demographic category, group, industry, socioeconomic status, etc. for some time. Many of these factors are accounted for in insurer's rating manuals. Studies published over the years by Wennberg, et al.,1 illustrated in the Dartmouth Atlas, have pointed out the small area variability of costs in the Medicare program, even when controlling for differences in illness burden. In the Wennberg studies, the variation is often connected to excess supply of hospital beds and specialists, and higher cost is often associated with lower quality and does not necessarily produce better outcomes. A recent 2004 Web edition of Health Affairs2 has a number of papers related to these issues.

The Specifics

For many years, I have been speaking and writing about variations in efficiency and pricing of hospital services at the provider specific level. These provider specific variations lead to the small area variations pointed out by the Dartmouth Atlas Group and others. One of the papers in the Health Affairs Web edition, cited above, also focuses on the provider specific variations of Academic Medical Centers. These excess costs, associated with lower quality and no better outcomes, are among the driving factors in the high cost of health care.

At the 2004 SOA Spring Meeting in Anaheim, I spoke about provider specific variability. Exhibit A summarizes the results from one of my slides on hospital in-patient variability. This data represents an analysis of 17 of the larger hospitals in a large metropolitan area, drawn from my standardized analysis of virtually all hospitals. This exhibit compares relative quality, charge, length- of-stay efficiency, and estimated reimburse-ment differences between the various hospitals. These results are all adjusted for case-mix and severity to put the analysis on an equitable basis.

The analysis behind Exhibit A is based on publicly available Medicare In-patient Data (MEDPAR). This was used because it is the most consistent and comprehensive source available, and is subject to standardized audit procedures. Also done were statistical studies that showed very high correlations between Medicare and commercial charge levels and LOS efficiency (percentage of days avoidable compared to most efficient practice benchmarks). The Wennberg article1 cites similar consistency between Medicare and commercial experience. With this background, carrier's discounts were applied for these providers to get their estimated reimbursement relativities.

The first column anonymously identifies the hospital (each one is known in the actual data). The second column identifies their quality scores on four patient-safety measures from the Agency for Health Research and Quality (AHRQ) developed for the federal agency by Stanford University/U.C. San Francisco. Each L represents one of the four categories where the hospital fell in the lower 15th percentile (unfavorable) on a quality measure. Each H represents results in the higher 15th percentile (favorable). All others not so designated fell between these ranges. Thus, an A in column two (Quality Measure) represents all four results in the middle range, e.g., 1L represents 1L and 3A's.

The third column represents the relative charge levels per day, all indexed to hospital K as 100 percent. Column four represents the percentage of med-surg days avoidable based on our LOS Efficiency™ Analysis. This is based on step-wise regression models considering all diagnoses, procedures, demographics, admit source and discharge disposition at the case-type/severity level.

The last column applies the efficiency and the carrier's discounts to determine the relative reimbursement level to each hospital. All the results from the table are case-mix severity adjusted to an equivalent basis.

In this example, results show the lower quality hospitals tend to be less efficient and get higher relative reimbursement. Thus, the marketplace is in effect rewarding in-efficiency and lower quality. Bringing the higher cost/less efficient hospital re-imbursement down to the average level of hospital K, L, P and Q, while improving quality to comparable levels, could substantially reduce medical costs in this geographic area while increasing overall quality.

A Sense of Commonality

This kind of variability is common across most geographic areas. The problem we face is to translate this information on variability into information that is actionable and will improve these patterns in the future. This will require the dissemination of provider comparative data on costs, quality, efficiency and efficacy of care. This information must be integrated into the proper incentive/disincentive structures within benefit programs in order to move patients to more cost effective/efficacious venues. This will force the offending providers to improve or risk failure.

The analysis of provider differences is well within the purview of actuarial applications, and the profession should be doing more in this field. However, our practical ex-perience with product design, underwriting, provider contracting and reimbursement put the profession squarely in the position of designing the practical applications of this analysis in the marketplace.

John P. Cookson, FSA, MAAA, is a consulting actuary for Milliman Inc.

1John E. Wennberg, et al. Use of Medicare Claims Data to Monitor Provider-Specific Performance Among Patients With Severe Chronic Illness.

2Health Affairs, Web Exclusive, October 7, 2004, Collection of 20 papers on variations in health care practice patterns and outcomes.