Preprint
Machine Learning

Bowel Outcome Prediction After Traumatic Spinal Cord Injury: Longitudinal Cohort Study

Chiara Pavese(Balgrist University Hospital, University of Zürich, Zürich, Switzerland), Lucas M. Bachmann(Medignition Inc., Research Consultants, Zürich, Switzerland), Martin Schubert(Balgrist University Hospital, University of Zürich, Zürich, Switzerland), Armin Curt(Balgrist University Hospital, University of Zürich, Zürich, Switzerland), Ulrich Mehnert(Balgrist University Hospital, University of Zürich, Zürich, Switzerland), Marc P. Schneider(Balgrist University Hospital, University of Zürich, Zürich, Switzerland), Giorgio Scivoletto(IRCCS Fondazione Santa Lucia, Rome, Italy), Enrico Finazzi Agrò(Tor Vergata University, Rome, Italy; Policlinico Tor Vergata, Rome, Italy), Doris Maier(BG-Trauma Center, Murnau, Germany), Rainer Abel(Hohe Warte, Bayreuth, Germany), Norbert Weidner(Heidelberg University Hospital, Heidelberg, Germany), Rüdiger Rupp(Heidelberg University Hospital, Heidelberg, Germany), Alfons G. Kessels(Maastricht University Medical Center, Maastricht, the Netherlands), Thomas M. Kessler(Balgrist University Hospital, University of Zürich, Zürich, Switzerland)
August 27, 2019Neurorehabilitation and Neural Repair21 citations
DOI

21

Citations

1

Influential Citations

Neurorehabilitation and Neural Repair

Venue

2019

Year

Abstract

Background. Predicting functional outcomes after traumatic spinal cord injury (SCI) is essential for counseling, rehabilitation planning, and discharge. Moreover, the outcome prognosis is crucial for patient stratification when designing clinical trials. However, no valid prediction rule is currently available for bowel outcomes after a SCI. Objective. To generate a model for predicting the achievement of independent, reliable bowel management at 1 year after traumatic SCI. Methods. We performed multivariable logistic regression analyses of data for 1250 patients with traumatic SCIs that were included in the European Multicenter Study about Spinal Cord Injury. The resulting model was prospectively validated on data for 186 patients. As potential predictors, we evaluated age, sex, and variables from the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI) and the Spinal Cord Independence Measure (SCIM), measured within 40 days of the injury. A positive outcome at 1 year post-SCI was assessed with item 7 of the SCIM. Results. The model relied on a single predictor, the ISNCSCI total motor score—that is, the sum of muscle strengths in 5 key muscle groups in each limb. The area under the receiver operating characteristics curve (aROC) was 0.837 (95% CI: 0.815-0.859). The prospective validation confirmed high predictive power: aROC = 0.817 (95% CI: 0.754-0.881). Conclusions. We generated a valid model for predicting independent, reliable bowel management at 1 year after traumatic SCI. Its application could improve counseling, optimize patient-tailored rehabilitation planning, and become crucial for appropriate patient stratification in future clinical trials.

Analysis

Why This Paper Matters

Predicting functional outcomes after traumatic spinal cord injury is critical for patient counseling, rehabilitation planning, and clinical trial design. Bowel management is a key aspect of quality of life, yet no validated prediction rule existed prior to this work. This paper addresses that gap by developing a simple, single-predictor model that can be easily applied in clinical settings.

The study leverages a large, multicenter European cohort (1250 patients) and validates the model prospectively on an independent sample (186 patients), lending credibility to its generalizability. The focus on a single, objective measure (ISNCSCI total motor score) makes the model practical and reproducible.

Technical Contributions

  • Single-predictor model: The final model uses only the ISNCSCI total motor score, which is the sum of muscle strengths in 5 key muscle groups per limb. This simplicity enhances clinical utility.
  • Robust validation: Prospective validation on a separate cohort confirms the model's discriminative power (aROC 0.817), reducing overfitting concerns.
  • Logistic regression approach: While not a novel machine learning technique, the use of multivariable logistic regression is appropriate for binary outcome prediction and provides interpretable odds ratios.

Results

  • Derivation cohort (n=1250): aROC = 0.837 (95% CI: 0.815-0.859)
  • Validation cohort (n=186): aROC = 0.817 (95% CI: 0.754-0.881)
  • The model outperforms chance and provides clinically meaningful discrimination, though no direct comparisons to other models are made.

Significance

This work provides a validated, easy-to-use tool for predicting bowel outcomes after SCI, which can directly impact patient care and trial design. For the AI community, it demonstrates the value of simple, interpretable models in high-stakes medical prediction tasks. The study underscores that complex models are not always necessary; a single well-chosen feature can yield strong predictive performance when grounded in domain knowledge.

logistic-regressionspinal-cord-injuryoutcome-predictionclinical-modelprognosisrehabilitation

Related Papers