Differences in methodology impact estimates of survival and dependence on home parenteral support of patients with nonmalignant short bowel syndrome

BACKGROUND: In patients with intestinal failure (IF), who are receiving home parenteral support (HPS), variations between centers in estimates of survival and HPS dependency often reflect differences in population characteristics. However, variations in methodology and adherence to model assumptions may further contribute.

OBJECTIVES: We investigated how differences in methodology affect estimates of outcomes in IF patients.

METHODS: We applied different model assumptions and statistical methods to real-life outcome data from a well-characterized cohort of nonmalignant short bowel syndrome (SBS) patients. This retrospective study was based on extracts from the Copenhagen IF database and from the Danish death registry.

RESULTS: Estimates of mortality varied substantially, depending on the study design and statistical method. The 5-y mortality rate obtained with the Kaplan-Meier (KM) method was estimated to be 10.1% higher if patients were only followed during HPS treatment, compared with follow-up regardless of HPS treatment. The 5-y cumulative incidence of weaning off HPS was overestimated by 4.4% when inappropriately using the KM method, instead of the cumulative incidence function. The 5-y survival rates in nonmalignant SBS-IF patients who initiated HPS were 89.1% for those aged younger than 40 y, 74.8% for patients aged 40-60 y, and 52.1% for those older than 60 y. A Cox regression analysis identified age and diagnoses other than inflammatory bowel disease as significant risk factors for mortality. For HPS dependency, bowel anatomy was significantly associated with the ability to wean off, and no patients without a colon and less than 100 cm remnant of the small bowel remained continuously weaned off and alive for one year.

CONCLUSIONS: The large variations in outcomes illustrated in this study emphasize the importance of the appropriate selection of statistical methods. A comparison between studies is problematic, due to differences in the methods employed.