Who is not coming to clinic? A predictive model of excessive missed appointments in persons with multiple sclerosis.

BACKGROUND

Missed appointments can have negative effects on several facets of healthcare, including disruption of services, worse patient health outcomes, and increased costs. The influence of demographic and clinical factors on missed appointments has been studied in a number of chronic conditions, but not yet in multiple sclerosis (MS). Engagement in healthcare services is a particular concern with this population, given the complexity of the condition. Furthermore, excessive missed appointments has emerged as a risk factor for suboptimal adherence to disease modifying therapies (DMTs), prompting further exploration into this issue and whether a tool could be developed to triage possible interventions for persons with MS on DMTs who are missing their appointments. As such, this study aimed to investigate the rate and factors associated with missed appointments among a large national sample of persons with MS and develop a predictive model of excessive missed appointments.

METHODS

Administrative data from 01/01/2013 to 12/31/2015 were extracted from the VA MS Center of Excellence Data Repository. Variables not related to excessive missed appointments, defined as missing more than 20% of scheduled appointments, in bivariate analyses (p > 0.20) were excluded. Remaining baseline co-occurring conditions, demographic, and healthcare utilization variables were entered into a logistic regression model, using a backward elimination criteria of p < 0.05. Calibration and discrimination of the model were assessed. An initial predictive score was generated based on the value of the variable and its β-value from the final model.

RESULTS

The number of missed appointments ranged from 0 to 84 over a two-year period. Over 59% missed at least one appointment, though only 4.28% had excessive missed appointments. Seven variables were retained in the model: adherence to DMTs, age, distance, histories of post-traumatic stress disorder, congestive heart failure, and chronic obstructive pulmonary disease, and emergency visits. Predictive scores ranged from -6.42 to 0.96 (M = -2.61, SD = 1.15). The final model had good discrimination, calibration, and fit.

CONCLUSIONS

By using this model and accompanying score, clinicians could have a good chance of predicting individuals who will miss more than 20% of their appointments and triaging interventions.