Key financial ratios can foretell hospital closures

Healthcare Financial Management, Nov, 1993 by Monty L. Lynn, Paul Wertheim

The final statistical step in testing individual ratios was to calculate the degree to which a ratio could accurately foretell a hospital's operating status. Logit analysis was used for this determination. The prediction percentage of hospitals correctly identified as open or closed is shown in the third and fifth columns of Exhibit 2.

The most predictive individual variable one year prior to closure is the profitability ratio of net income to total revenues, which has a highly significant chi-square value of 22.9 and a predictive accuracy of TABULAR DATA OMITTED 69.2 percent. (The predictive accuracy percentage suggests that 69.2 percent or 98 of the sample hospitals were correctly identified as open or closed.) Two years before closure, the viability index is the most predictive with a chi-square value of 9.4 and a predictive accuracy of 61.2 percent.

Ratios combined

The next aspect of the indicators to be examined was whether factoring several ratios together would yield an even better prediction of a hospital's financial status. Seven hundred possible combinations of ratios were tested, and the model with the highest predictability one year and two years prior to closure was determined.

To avoid statistical problems associated with variables that are highly intercorrelated and to keep the model more practical for industry users, only one ratio from each of the four categories was included in any given model. Thus, each model contained one leverage ratio, one liquidity ratio, one capital efficiency ratio and one resource availability ratio.

The multivariate model with the highest predictive accuracy (75.0 percent one year prior to closure and 73.8 percent two years prior to closure) and the highest chi-square value (43.7, one year; 25.1, two years) is total liabilities/total assets (leverage) total assets/current liabilities (liquidity) total revenue/total expenses (capital efficiency) total assets/bed days available (resource availability). Interpreted, the 73.8 percent two-year prediction accuracy indicates that the status (open or closed) of 105 of the 142 sample hospitals is correctly identified with financial data taken two years prior to closure.

When compared to the most predictive individual ratios from Exhibit 2, the combined model provides an additional accuracy of 5.8 percent to 12.6 percent in predicting hospital status. A prediction accuracy in the 73 percent to 75 percent range, as found here, is especially strong when considering that the decision to close a not-for-profit hospital is often influenced by many nonfinancial considerations. Additionally, recalling that the model is differentiating between open and closed hospitals that are matched in bed capacity, urban/rural location, and state location, suggests further that a correct prediction of closed or open status is fairly fine-tuned.

After the most predictive model was identified, one final statistical test was made. By comparing the most predictive model with other combinations of variables, the incremental explanatory power of TABULAR DATA OMITTED each of the ratios in the model could be measured. The F-tests, shown in Exhibit 3, indicate how much explanatory power an individual variable adds to the multi-variate model. The higher the F-ratio, the more an individual financial ratio contributes to the model's predictive accuracy.