Volume 16, Number 9—September 2010
Long-Term Health Risks for Children and Young Adults after Infective Gastroenteritis
|Type of sequelae||First-time hospitalizations
||Crude rate ratio, RR (95% CI)||Adjusted† rate ratio, RR (95% CI)||Adjusted AR, %‡||Goodness of fit§|
|Any||5,634||27.8||41,054||11.8||2.36 (2.28–2.41)||1.64 (1.59–1.67)||39||0.05|
|Intragastrointestinal||1,267||4.0||9,385||2.3||1.77 (1.67–1.88)||1.52 (1.42–1.62)||34||0.04|
|Extragastrointestinal||5,045||24.1||34,425||9.8||2.46 (2.39–2.54)||1.63 (1.57–1.68)||39||0.08|
*Rates are per 100,000 person-years. RR, relative risk; CI, confidence interval; AR, attributable risk.
†Multivariate Cox regression estimating the adjusted rate ratio of first-time hospitalization for any, intragastrointestinal, and extragastrointestinal sequelae. Adjusted for gender, indigenous status, year of birth, age at exposure or proxy, singleton, weight at birth, hospital birth, mother’s region of birth, father’s region of birth, socioeconomic status, accessibility to services and previous hospitalization for comorbidity.
‡Proportion of first-time hospitalizations for sequelae where previous exposure to an enteric infection was a component cause.
§Pseudo R2. As explained by Hosmer and Lemeshow (20), a measure analogous to R2 would be useful as a measure of Cox regression model performance; however, although a pseudo R2 can be calculated the values obtained are often low because of the censored nature of the data even though the model is adequate. In our models the R2 values were 0.05, 0.04 and 0.08 for the 3 models (any, intragastrointestinal, and extragastrointestinal), respectively. The models generated were population-based descriptive models, which aimed to evaluate the average effect on survival to first-time hospitalization with the outcome of interest adjusted for known and measurable confounders, rather than predict the probability of survival for a specified individual. Thus, the most important assessment criteria for evaluating the appropriateness of a descriptive Cox regression model is that the proportional hazards assumption is not violated and the overall model is significant. In all of our models the proportional hazards assumption was tested and found not to be violated and the overall model significance was Prob > χ2 <0.00005.