Development and validation of a clinical prediction rule for hospitalization due to pneumonia or influenza or death during influenza epidemics among community-dwelling elderly persons.

BACKGROUND: Uncertainties among health care providers and patients about the risk of serious influenza-associated complications and the potential benefits of vaccination may contribute to unsatisfactorily low influenza vaccination rates. To quantify the risk of serious outcomes (hospitalization due to pneumonia or influenza or death due to any cause) during influenza seasons, we developed a clinical prediction rule for the probability of hospitalization due to pneumonia or influenza or death among elderly persons.
METHODS: We developed the clinical prediction rule using data from linked administrative databases in a cohort of 16,280 noninstitutionalized and unvaccinated elderly persons. Validation of the rule was conducted in 5 unvaccinated and 6 vaccinated cohorts, each consisting of >11,000 elderly members of 3 managed care organizations. Logistic regression was used to produce a prognostic score on the basis of the following predictors: age; sex; presence of pulmonary, cardiac, and renal disease; dementia or stroke and cancer; number of outpatient visits; and hospitalization due to pneumonia or influenza during the previous year.
RESULTS: Reliability of the regression model was good (P=.65, by goodness-of-fit test), and it discriminated well between those who did and those who did not experience an outcome (area under the receiver-operating curve, 0.83; 95% confidence interval, 0.81-0.85). Validation revealed moderately lower but acceptable discriminating values (0.72-0.81). In the derivation cohort, the prognostic accuracy of the rule was high when a cutoff score for the upper 50th percentile was used: > or =10 of 1000 subjects with a score in the upper 50th percentile were predicted to have an outcome, and 89% of all outcomes were observed in this high-risk group, whereas <10 of 1000 subjects with a score in the lower 50th percentile were predicted to have an outcome, and only 11% of outcomes occurred in this group. Among unvaccinated subjects in the single-derivation cohort and the 11 validation cohorts combined, the outcome event rates were 35 events/1000 subjects in the higher-risk group and 6 events/1000 subjects in the lower-risk group. With vaccination, these event rates dropped by 15 events/1000 subjects and 2 events/1000 subjects, respectively.
CONCLUSIONS: This prediction rule may be a useful tool to complement other age-based strategies, to further encourage vaccination, especially among those at the highest risk of serious complications due to influenza.