A promise of machine learning in health care is the avoidance of biases in diagnosis and treatment; a computer algorithm could objectively synthesize and interpret the data in the medical record. Integration of machine learning with clinical decision support tools, such as computerized alerts or diagnostic support, may offer physicians and others who provide health care targeted and timely information that can improve clinical decisions. Machine learning algorithms, however, may also be subject to biases. The biases include those related to missing data and patients not identified by algorithms, sample size and underestimation, and misclassification and measurement error. There is concern that biases and deficiencies in the data used by machine learning algorithms may contribute to socioeconomic disparities in health care. This Special Communication outlines the potential biases that may be introduced into machine learning–based clinical decision support tools that use electronic health record data and proposes potential solutions to the problems of overreliance on automation, algorithms based on biased data, and algorithms that do not provide information that is clinically meaningful. Existing health care disparities should not be amplified by thoughtless or excessive reliance on machines.

Objective: We investigated potential interactions between human leukocyte antigen (HLA) genotype and body mass index (BMI) status in relation to the risk of developing multiple sclerosis (MS). Methods: We used 2 case-control studies, one with incident cases (1,510 cases, 2,017 controls) and one with prevalent cases (937 cases, 609 controls). Subjects with different genotypes and BMI were compared with regard to incidence of MS by calculating odds ratios (ORs) with 95% confidence intervals (CIs) employing logistic regression. Potential interactions between genotypes and BMI were evaluated by calculating the attributable proportion due to interaction. Results: In both cohorts, a significant interaction was observed between HLA-DRB1*15 and obesity, regardless of HLA-A*02 status. Similarly, there was a significant interaction between absence of A*02 and obesity, regardless of DRB1*15 status. In the incident cohort, obese subjects with the most susceptible genotype (carriage of DRB1*15 and absence of A*02) had an OR of 16.2 (95% CI 7.5–35.2) compared to nonobese subjects without the genetic risk factors. The corresponding OR in the prevalent study was 13.8 (95% CI 4.1–46.8). Conclusions: We observed striking interactions between BMI status and HLA genotype with regard to MS risk. Hypothetically, a low-grade inflammatory response inherent to obesity synergizes with the adaptive, HLA molecule–restricted arm of the immune system, causing MS. Prevention of adolescent obesity may thus lower the risk of developing MS, predominantly among people with a genetic susceptibility to the disease.

Objective: We sought to estimate the causal effect of low serum 25(OH)D on multiple sclerosis (MS) susceptibility that is not confounded by environmental or lifestyle factors or subject to reverse causality. Methods: We conducted mendelian randomization (MR) analyses using an instrumental variable (IV) comprising 3 single nucleotide polymorphisms found to be associated with serum 25(OH)D levels at genome-wide significance. We analyzed the effect of the IV on MS risk and both age at onset and disease severity in 2 separate populations using logistic regression models that controlled for sex, year of birth, smoking, education, genetic ancestry, body mass index at age 18–20 years or in 20s, a weighted genetic risk score for 110 known MS-associated variants, and the presence of one or more HLA-DRB1*15:01 alleles. Results: Findings from MR analyses using the IV showed increasing levels of 25(OH)D are associated with a decreased risk of MS in both populations. In white, non-Hispanic members of Kaiser Permanente Northern California (1,056 MS cases and 9,015 controls), the odds ratio (OR) was 0.79 (p = 0.04, 95% confidence interval (CI): 0.64–0.99). In members of a Swedish population from the Epidemiological Investigation of Multiple Sclerosis and Genes and Environment in Multiple Sclerosis MS case-control studies (6,335 cases and 5,762 controls), the OR was 0.86 (p = 0.03, 95% CI: 0.76–0.98). A meta-analysis of the 2 populations gave a combined OR of 0.85 (p = 0.003, 95% CI: 0.76–0.94). No association was observed for age at onset or disease severity. Conclusions: These results provide strong evidence that low serum 25(OH)D concentration is a cause of MS, independent of established risk factors.

Objective: To utilize Mendelian randomization to estimate the causal association between low serum vitamin D concentrations, increased body mass index (BMI), and pediatric-onset multiple sclerosis (MS) using genetic risk scores (GRS). Methods: We constructed an instrumental variable for vitamin D (vitD GRS) by computing a GRS for 3 genetic variants associated with levels of 25(OH)D in serum using the estimated effect of each risk variant. A BMI GRS was also created that incorporates the cumulative effect of 97 variants associated with BMI. Participants included non-Hispanic white individuals recruited from over 15 sites across the United States (n = 394 cases, 10,875 controls) and Sweden (n = 175 cases, 5,376 controls; total n = 16,820). Results: Meta-analysis findings demonstrated that a vitD GRS associated with increasing levels of 25(OH)D in serum decreased the odds of pediatric-onset MS (odds ratio [OR] 0.72, 95% confidence interval [CI] 0.55, 0.94; p = 0.02) after controlling for sex, genetic ancestry, HLA-DRB1*15:01, and over 100 non–human leukocyte antigen MS risk variants. A significant association between BMI GRS and pediatric disease onset was also demonstrated (OR 1.17, 95% CI 1.05, 1.30; p = 0.01) after adjusting for covariates. Estimates for each GRS were unchanged when considered together in a multivariable model. Conclusions: We provide evidence supporting independent and causal effects of decreased vitamin D levels and increased BMI on susceptibility to pediatric-onset MS.