Natural language processing of referral letters improves machine learning based triaging of patients with low back pain to the most appropriate intervention. (Preprint)

BACKGROUND Decision-support systems for the suggestion of optimal treatments for individual patients with back pain are currently insufficiently accurate to use for clinical application. Most of the input provided to train these systems is based on patient reported outcome measures. However, with the appearance of electronic patient records, additional qualitative data on reasons for referrals, and patients’ goals become available for support systems. There are currently no decision support tools that cover a wide range of biopsychosocial factors including referral letter information to help clinicians triage patients to the optimal LBP-treatment. OBJECTIVE The objective of this study was to investigate the added value of including qualitative data from electronic patient records on the accuracy of a quantitative decision-support system for patients with low back pain. METHODS A retrospective study in a clinical cohort of Dutch patients with low back pain was carried out. Patients filled out a baseline questionnaire about demographics, pain, disability, work status, quality of life, medication, psychosocial functioning, comorbidity, history and duration of pain. Referral reasons and patient request for help (patient goals) were extracted via natural language processing (NLP) and enriched the dataset. For decision support, these data were considered as independent factors for the choice to triage to neurosurgery, anesthesiology, rehabilitation or minimal intervention. Support Vector Machine, Boosted Decision Tree and Multiclass Neural Network (NN) models are trained for two conditions: with and without consideration of referral letter content. The models’ accuracies were evaluated via F1 Scores, and Confusion Matrices to predict treatment-path (out of 4 paths) with and without the additional referral parameters. RESULTS Data of 1608 patients were evaluated. The evaluation indicates that two referral reasons from the referral letter (for an anesthesiology and a rehabilitation intervention) increased the F1-Score accuracy by up to 19.5% for the triaging. The results were confirmed by the confusion matrices. CONCLUSIONS This study indicates that data-enriching by adding NLP-based extraction of the content of referral letters increases the model accuracy of decision support systems in suggesting optimal treatments for individual patients with low back pain. Overall model accuracies were considered low and insufficient for clinical application.