Detaillierte Beitrags-Information

Zurück zur Liste

Beitragstitel Worse postoperative outcomes for "drop-outs", as predicted by machine-learning based model.
Beitragscode P16
Autoren
  1. Anne F. Mannion Schulthess Clinic, Zurich, Switzerland Vortragender
  2. Daniel Müller
  3. Dave O'Riordan
  4. Tamas F. Fekete Schulthess Klinik Zürich
  5. François Porchet Schulthess Klinik Zürich
  6. Frank Stefan Kleinstück Schulthess Klinik Zürich
  7. Raluca Reitmeir Schulthess Klinik
  8. Markus Loibl Schulthess Klinik
  9. Dezsö Jeszenszky Schulthess Klinik Zürich
  10. Daniel Haschtmann Schulthess Clinic Zurich
Präsentationsform Poster
Themengebiete
  • A03 - Wirbelsäule
Abstract INTRODUCTION
Lack of compliance with follow-up can threaten the validity of outcomes reported in registries and clinical studies. We used a published outcome-predictor model to predict (based on baseline characteristics) the outcomes of patients failing to return a questionnaire 12 months after surgery (dropouts) and compare them with the outcomes of those that did return a questionnaire (remainers).
METHODS
The data of all patients with degenerative spinal disorders of the thoracic or lumbar spine included in our in-house spine outcomes registry from 1.1.2006 to 31.12.2017 were analysed. Using the data of remainers (8374/9189 (91%) cases), a model was developed to predict multidimensional outcome 12 months after surgery (COMI, and its domain scores) based on 20 key baseline variables (Müller et al 2021). This was used to predict the outcome of both dropouts and remainers. The groups were compared using unpaired t-tests.
RESULTS
The predicted outcome scores of the remainers did not differ significantly (p>0.05) from their actual outcome scores, suggesting the model was sufficiently accurate. The mean baseline scores of dropouts were significantly (p < 0.05) worse than those of remainers for most domains. The predicted outcome of dropouts at 12mo FU was significantly (p< 0.05) worse than that of the remainers for all domains.
DISCUSSION
Dropouts at follow-up introduced a significant and consistent bias in reported outcomes. Although the size of the effect was small in this particularly compliant cohort (with 91% follow-up at 12mo), when evaluating healthcare providers with poorer follow-up rates such a bias may be sufficient to threaten valid comparisons. The bias would overestimate the performance of hospitals with lower follow-up rates (perhaps also failing to detect poorly performing hospitals) and underestimate that of hospitals with high follow-up rates. If using spine surgery registries to perform benchmarking activities, the difference in follow-up rates between hospitals must be considered and adjusted for.