Generalisable prediction models for outcomes after lumbar spinal stenosis surgery: a model development and external validation study

Bjørnar Berg; Allan Abbott; Casper Friis Pedersen; Henrik Hedevik; Martin A Gorosito; Henrik Lykke Joakimsen; Per Joel Burman; Karl Øyvind Mikalsen; Mikkel Østerheden Andersen; Tor Ingebrigtsen; Tore Solberg; Margreth Grotle

doi:10.1016/j.eclinm.2026.103989

Generalisable prediction models for outcomes after lumbar spinal stenosis surgery: a model development and external validation study

EClinicalMedicine. 2026 May 28:96:103989. doi: 10.1016/j.eclinm.2026.103989. eCollection 2026 Jun.

Authors

Bjørnar Berg¹, Allan Abbott^{2

3}, Casper Friis Pedersen⁴, Henrik Hedevik², Martin A Gorosito^{1

5}, Henrik Lykke Joakimsen^{6

7}, Per Joel Burman⁷, Karl Øyvind Mikalsen^{6

7

8}, Mikkel Østerheden Andersen⁴, Tor Ingebrigtsen^{6

9}, Tore Solberg^{6

9}, Margreth Grotle^{1

10}

Affiliations

¹ Centre for Intelligent Musculoskeletal Health, Faculty of Health Sciences, Oslo Metropolitan University, Oslo, Norway.
² Unit of Physiotherapy, Division of Prevention, Rehabilitation and Community Medicine, Department of Health, Medicine and Caring Sciences, Linköping, Sweden.
³ Clinical Department of Orthopaedics in Linköping, Region Östergötland, Linköping, Sweden.
⁴ Center for Spine Surgery and Research, Spine Center of Southern Denmark, Lillebaelt Hospital, University of Southern Denmark, Kolding, Denmark.
⁵ Department of Computer Science, Oslo Metropolitan University, Oslo, Norway.
⁶ Department of Clinical Medicine, Faculty of Health Sciences, UiT The Arctic University of Norway, Tromsø, Norway.
⁷ The Norwegian Centre for Clinical Artificial Intelligence, University Hospital of North Norway, Tromsø, Norway.
⁸ UiT Machine Learning Group, Department of Physics and Technology, UiT the Arctic University of Norway, Tromso, Norway.
⁹ Department of Neurosurgery and The Norwegian Registry for Spine Surgery, The University Hospital of North Norway, Tromsø, Norway.
¹⁰ Division of Clinical Neuroscience, Department of Research and Innovation, Oslo University Hospital, Oslo, Norway.

Abstract

Background: Lumbar spinal stenosis is a leading indication for spine surgery, but outcomes are heterogeneous. We aimed to develop and externally validate prediction models for 12-month disability and pain to inform shared decision-making.

Methods: This registry-based multicentre cohort study used data from three national spine registries of patients (≥16 years) undergoing elective lumbar spinal stenosis surgery. Data from the Norwegian Registry for Spine Surgery (NORspine, 2007-2023) were used for model development and internal-external cross-validation (IECV). External validation was carried out in the Swedish Registry (SweSpine, 2016-2022) and Danish Registry (DaneSpine, 2009-2022) with data collected by the Spine Centre of Southern Denmark. The primary outcome was the Oswestry Disability Index (ODI) at 12 months, modelled as a continuous and binary measure (acceptable symptom state). Secondary outcomes were Numeric Rating Scale (NRS) back and leg pain at 12 months. Logistic regression, linear regression, and XGBoost models were applied with 16 predictors. Missing data were handled using multiple imputation. Performance was assessed by calibration, mean absolute error (MAE), adjusted R², and C-statistics. This study is registered with Open Science Framework (https://osf.io/qz27b/).

Findings: The development cohort included 31,908 patients (52.4% female, 47.6% male). The external validation cohorts included 30,700 from SweSpine (52.8% female, 47.2% male) and 4063 from DaneSpine (54.6% female, 45.4% male). Twelve-month outcome completeness was 77% in the development cohort and ranged from 66% to 80% across the external validation cohorts. For ODI, linear regression achieved a pooled MAE of 12.4 (95% CI 11.8-13.1) after IECV, and 13.3 (95% CI 13.2-13.4) and 12.3 (95% CI 12.0-12.7) at external validation. Adjusted R² values ranged from 0.26 to 0.33. Calibration was acceptable, with slopes near 1 and calibration-in-the-large ranging from -0.47 after IECV to 1.28-1.54 at external validation, indicating minor systematic underprediction. The binary ODI model achieved C-statistics of 0.75 (95% CI 0.74-0.76) after IECV, and 0.78 (95% CI 0.78-0.79) and 0.76 (95% CI 0.74-0.77) at external validation. Pain models showed lower performance (MAE 2.2-2.6; C-statistics 0.64-0.73). XGBoost yielded similar results.

Interpretation: Models predicting disability and pain were well calibrated and generalisable across Scandinavian countries, with the best overall performance for disability. These findings provide a foundation for prospective evaluation in future studies to determine the impact on decision-making and patient outcomes in clinical practice.

Funding: Research Council of Norway.

Keywords: External validation; Lumbar spinal stenosis; Machine learning; Prediction model; Spine surgery.