EP638 - Utility of State Transition Models As A Measure of Intestinal Recovery

Purpose/Background: Flatus, bowel movements, and reduction in hospital length of stay (HLOS) are often used as outcome measures when studying the effectiveness of interventions intended to accelerate intestinal recovery. However, these measures can be imprecise, subjective, and have low statistical power, resulting in the need to enroll numerous patients in clinical trials which use intestinal recovery as a study endpoint. We propose using Bayesian longitudinal ordinal state transition models (BLOSTM) to overcome these limitations.

Methods/Interventions: A synthetic dataset was created with 100 patients in two groups that had the same characteristics of a cohort of 29 patients who were enrolled in an RCT of interventions intended to reduce HLOS for small bowel obstruction. Groups were created to either show minimal difference in outcomes, or a large difference in outcomes. Patients were assigned one of 5 clinical states for every day in a 31-day period following surgery: 1-at home, 2-in hospital, eating, 3- in hospital, NPO, 4- in hospital- NGT in place, or 5- Having complications, surgery or death. States 1 and 2 were considered favorable states and States 3, 4, and 5 as unfavorable states. The probability of being in or transitioning from one state to another for every observation day was calculated. The number of days benefited from being in a favorable versus unfavorable state from an intervention was calculated.

Results/Outcomes: BLOSTM of the synthetic data with minimal differences in outcomes between groups is shown in the figure panels on the left, and the distribution of probabilities in both transitioning of state and number of days benefited was flat, consistent with no differences between groups. The panels on the right demonstrate results for the synthetic data created to have a beneficial effect of an intervention intended to accelerate intestinal recovery. There is a distinct peak in the probability distribution, and patients in the intervention group had 22 more days (out of 31 days) of being in a favorable relative to unfavorable clinical state.

Conclusion/Discussion: Easily collected daily clinical status data can be used to determine differences between groups in intestinal recovery trials. Because this is a Bayesian analysis, sample size and problems related to multiple comparison do not exist so that data can be collected and analyzed as many times as desired as an experiment proceeds. The statistical power for this analysis is substantially greater than simple time to flatus or hospital discharge, requiring many fewer patients for clinical trials to yield statistical significance.

Probability Distribution of Mean Days Recovered and Days Benefited With Synthetic Groups Aimed to Demonstrate Minimal Difference in Outcomes
The probability distribution is flat and centered about 0, demonstrating no difference in outcomes
Probability Distribution of Mean Days Recovered and Days Benefited With Synthetic Groups Aimed to Demonstrate a Large Difference in Outcomes
The probability distribution is peaked, with the intervention group demonstrating 22 more days (out of 31 days) of being in a favorable state