Artificial Intelligence Enabled Decision Support for Selection of Patients for Lumbar Spine Surgery

Background One third of patients operated for lumbar disc herniation (LDH) or spinal stenosis (LSS) do not achieve substantial improvement. Studies indicate that well informed shared decision making (SDM) can improve the selection to surgery, and thus the outcomes. Numerous algorithms for outcome prediction have therefore been developed, and some use artificial intelligence (AI). Most are trained on small datasets, few are accurate, all are stand-alone or web-based applications not integrated in the electronic health record (EHR), and none are implemented in routine clinical practice.

The Norwegian registry for spine surgery (NORspine) comprises a cohort of more than 69,000 cases. The investigators have used AI to analyze the dataset and predict the outcome, and developed a decision support tool (DST) which is seamlessly integrated in the EHR DIPS Arena®.

The investigators intend to use the tool to inform the SDM between surgeons and patients about the indication for surgery (yes or no), to increase the proportion with a successful outcome. The aim of the study is to assess the safety and feasibility of the DST for use in a subsequent pilot study.

The device The DST (the device) is an integrate compound of software-solutions. Baseline data are registered by patients and surgeons on questionnaires integrated in DIPS Arena®, and transferred to NORspine. The data are also transferred (de-identified) to the AI-enabled prediction algorithm which operates in a cloud-based model hosting service. The algorithm has been trained and validated on a dataset from NORspine. The area under the curve for prediction of the main outcome (Oswestry disability index after12 months) in receiver operating characteristic analysis is very high (0.85) for LDH and moderate (0.72) for LSS. The model host also calculates outcomes (proportions with substantial, slight, or no improvement, and worsening) for the 50 cases with baseline variables most similar to the present case ("patients-like-me"). Finally, the individual prediction and the outcomes for the "patients-like-me" are transferred back and displayed in the regular user interface of DIPS Arena® for use in the SDM.

Clinical investigations For this feasibility study, the investigators will use convergent qualitative and quantitative mixed methods. The comparator is decision making in routine clinical practice, without use of the DST. The study will include 20 patients with magnetic resonance imaging confirmed LDH or LSS referred for evaluation of the indication for surgery, and six surgeons who do the evaluations. The study will iteratively redesign the user interface of the DST until it is considered safe and feasible for use in a following pilot study.