Relative Performance of Machine Learning and Linear Regression in Predicting Quality of Life and Academic Performance of School Children in Norway: Data Analysis of a Quasi-Experimental Study.

BACKGROUND: Machine learning techniques are increasingly being applied in health research. It is not clear how useful these approaches are for modeling continuous outcomes. Child quality of life is associated with parental socioeconomic status and physical activity and may be associated with aerobic fitness and strength. It is unclear whether diet or academic performance is associated with quality of life. OBJECTIVE: The purpose of this study was to compare the predictive performance of machine learning techniques with that of linear regression in examining the extent to which continuous outcomes (physical activity, aerobic fitness, muscular strength, diet, and parental education) are predictive of academic performance and quality of life and whether academic performance and quality of life are associated. METHODS: We modeled data from children attending 9 schools in a quasi-experimental study. We split data randomly into training and validation sets. Curvilinear, nonlinear, and heteroscedastic variables were simulated to examine the performance of machine learning techniques compared to that of linear models, with and without imputation. RESULTS: We included data for 1711 children. Regression models explained 24% of academic performance variance in the real complete-case validation set, and up to 15% in quality of life. While machine learning techniques explained high proportions of variance in training sets, in validation, machine learning techniques explained approximately 0% of academic performance and 3% to 8% of quality of life. With imputation, machine learning techniques improved to 15% for academic performance. Machine learning outperformed regression for simulated nonlinear and heteroscedastic variables. The best predictors of academic performance in adjusted models were the child's mother having a master-level education (P<.001; ß=1.98, 95% CI 0.25 to 3.71), increased television and computer use (P=.03; ß=1.19, 95% CI 0.25 to 3.71), and dichotomized self-reported exercise (P=.001; ß=2.47, 95% CI 1.08 to 3.87). For quality of life, self-reported exercise (P<.001; ß=1.09, 95% CI 0.53 to 1.66) and increased television and computer use (P=.002; ß=-0.95, 95% CI -1.55 to -0.36) were the best predictors. Adjusted academic performance was associated with quality of life (P=.02; ß=0.12, 95% CI 0.02 to 0.22). CONCLUSIONS: Linear regression was less prone to overfitting and outperformed commonly used machine learning techniques. Imputation improved the performance of machine learning, but not sufficiently to outperform regression. Machine learning techniques outperformed linear regression for modeling nonlinear and heteroscedastic relationships and may be of use in such cases. Regression with splines performed almost as well in nonlinear modeling. Lifestyle variables, including physical exercise, television and computer use, and parental education are predictive of academic performance or quality of life. Academic performance is associated with quality of life after adjusting for lifestyle variables and may offer another promising intervention target to improve quality of life in children.

Responsiveness, Reliability, and Minimally Important and Minimal Detectable Changes of 3 Electronic Patient-Reported Outcome Measures for Low Back Pain: Validation Study.

BACKGROUND: The Roland Morris Disability Questionnaire (RMDQ), visual analog scale (VAS) of pain intensity, and numerical rating scale (NRS) are among the most commonly used outcome measures in trials of interventions for low back pain. Their use in paper form is well established. Few data are available on the metric properties of electronic counterparts. OBJECTIVE: The goal of our research was to establish responsiveness, minimally important change (MIC) thresholds, reliability, and minimal detectable change at a 95% level (MDC95) for electronic versions of the RMDQ, VAS, and NRS as delivered via iOS and Android apps and Web browser. METHODS: We recruited adults with low back pain who visited osteopaths. We invited participants to complete the eRMDQ, eVAS, and eNRS at baseline, 1 week, and 6 weeks along with a health transition question at 1 and 6 weeks. Data from participants reporting recovery were used in MIC and responsiveness analyses using receiver operator characteristic (ROC) curves and areas under the ROC curves (AUCs). Data from participants reporting stability were used for analyses of reliability (intraclass correlation coefficient [ICC] agreement) and MDC95. RESULTS: We included 442 participants. At 1 and 6 weeks, ROC AUCs were 0.69 (95% CI 0.59 to 0.80) and 0.67 (95% CI 0.46 to 0.87) for the eRMDQ, 0.69 (95% CI 0.58 to 0.80) and 0.74 (95% CI 0.53 to 0.95) for the eVAS, and 0.73 (95% CI 0.66 to 0.80) and 0.81 (95% CI 0.69 to 0.92) for the eNRS, respectively. Associated MIC thresholds were estimated as 1 (0 to 2) and 2 (-1 to 5), 13 (9 to 17) and 7 (-12 to 26), and 2 (1 to 3) and 1 (0 to 2) points, respectively. Over a 1-week period in participants categorized as "stable" and "about the same" using the transition question, ICCs were 0.87 (95% CI 0.66 to 0.95) and 0.84 (95% CI 0.73 to 0.91) for the eRMDQ with MDC95 of 4 and 5, 0.31 (95% CI -0.25 to 0.71) and 0.61 (95% CI 0.36 to 0.77) for the eVAS with MDC95 of 39 and 34, and 0.52 (95% CI 0.14 to 0.77) to 0.67 (95% CI 0.51 to 0.78) with MDC95 of 4 and 3 for the eNRS. CONCLUSIONS: The eRMDQ was reliable with borderline adequate responsiveness. The eNRS was responsive with borderline reliability. While the eVAS had adequate responsiveness, it did not have an attractive reliability profile. Thus, the eNRS might be preferred over the eVAS for measuring pain intensity. The observed electronic outcome measures' metric properties are within the ranges of values reported in the literature for their paper counterparts and are adequate for measuring changes in a low back pain population.

Development and feasibility of an intervention featuring individual supported work placements to aid return to work for unemployed people living with chronic pain.

BACKGROUND: Working in good jobs is associated with good health. High unemployment rates are reported in those disabled with musculoskeletal pain. Supported employment interventions work well for helping people with mental health difficulties to gain and retain employment. With adaptation, these may be useful for people with chronic pain. We aimed to develop and explore the feasibility of delivering such an adapted intervention. METHODS: We developed an intervention and recruited unemployed people with chronic pain from NHS pain clinics and employment services. We trained case managers to assess participants and match them to six-week work placements in the Midlands and provide ongoing support to them and their managers. Participants attended a two-day work preparation session prior to placement. Outcome measures included quality of life at baseline, six- weeks, 14-weeks, and six-months, and return to work at 14-weeks and six-months. We held focus groups or interviews with stakeholders to examine acceptability and experiences of the intervention. RESULTS: We developed an intervention consisting of work preparation sessions, work experience placements, and individualised employment support. We enrolled 31 people; 27 attended work preparation sessions, and 15 attended placements. Four of our participants started jobs during the study period. We are aware of two others starting jobs shortly after cessation of follow-up. We experienced challenges to recruitment in one area where we had many and diverse placement opportunities and good recruitment in another area where we had a smaller range of placement opportunities. All stakeholders found the intervention acceptable, and it was valued by those given a placement. While there was some disappointment among those not placed, this group still valued the work preparation sessions. CONCLUSIONS: The developed intervention was acceptable to participants and partners. Trialling the developed intervention could be feasible with attention to three main processes. To ensure advanced availability of a sufficiently wide range of work placements in each area, multiple partners would be needed. Multiple recruitment sites and focus on employment services will yield better recruitment rates than reliance on NHS pain clinics. Maintaining an adequate follow-up response rate will likely require additional approaches with more than the usual effort.