Risk for Medication Nonadherence Among Medicaid Enrollees With Fibromyalgia: Development of a Validated Risk Prediction Tool

Abstract

Objective To develop and validate a risk assessment tool called the Prescription Medication Non‐Adherence Prediction Tool (Rx‐NAPT) to predict medication nonadherence in patients with fibromyalgia. Methods This was a retrospective cohort study using claims data from South Carolina Medicaid. Patients with fibromyalgia who were ≥18 years old and who had filled at least 1 prescription medication for pregabalin, duloxetine, or milnacipran from January 1, 2005, through June 30, 2011 were included. Medication possession ratios (MPRs) were calculated to classify patients as adherent (MPR ≥ 80%) or nonadherent (MPR < 80%). Multivariable logistic models using 100 bootstrap replications (with replacement) were used to identify factors associated with medication nonadherence, including age, gender, race, days’ supply, medication type, and fibromyalgia‐related comorbidity score. Weighted β coefficients of the predictors were used to create the Rx‐NAPT. Youden's J statistic was used to classify nonadherent patients into different levels of risk. Results The study sample comprised 6,626 patients with fibromyalgia, where 4,804 (72.50%) were non‐adherent and 1,822 (27.50%) were adherent to their prescribed medication(s). Logistic regression models showed that 7 predictors (gender, age, race, fibromyalgia‐related comorbidity score, medication type, health maintenance organization coverage, emergency room visit) were statistically significant in ≥50% of the bootstrapped samples. The final model demonstrated reasonable discrimination (area under the curve [AUC] = 0.6224) and calibration (Hosmer‐Lemeshow goodness‐of‐fit; P > 0.05) statistics and was validated internally (AUC = 0.6372). Conclusion Poor adherence with medication remains an important barrier to providing optimal care. Our risk prediction model provides an easy tool to help clinicians better identify patients with fibromyalgia who may not take their medications as prescribed.

Data Science Methods for Nursing-Relevant Patient Outcomes and Clinical Processes

“…Other studies included clinical notes, 45,47,56,57 forum posts, 58 video/audio recordings, 42,44 and medical images 42,43 as their corpus of data. Study settings largely included patients and data from academic medical centers but also included community health clinics, 54 health Web sites, 58 and state Medicaid databases 49 . Outcomes and associated variables differed but often included cost of a service or clinical event (ie, readmission).…”

“…Study settings largely included patients and data from academic medical centers but also included community health clinics, 54 health Web sites, 58 and state Medicaid databases. 49 Outcomes and associated variables differed but often included cost of a service or clinical event (ie, readmission). Other outcomes included medication adherence, 49,53,58 fall prevention, 42 30-day morbidity, 51,52,59 service utilization, 55 and risk prediction.…”

“…45,47,[56][57][58] Studies used various data science methods to evaluate their outcomes. Several used some form of regression, 48,49,51,52,57,59 but the use of more advanced methods was prevalent as well. Of the 20 articles that addressed cost and utilization in this review, 5 used NLP, 45,47,[56][57][58] 5 used neural networks, 41,43,44,53,54 and 6 used other types of machine learning.…”

“…Many data sources included in these studies consisted of administrative data, which is often the primary source used to determine service cost. However, other studies collected data from patient surveys, 46,48 EHR data (clinical observations, clinical notes, and imaging), 42,43,46,[48][49][50][51][52][53][54] and smartphone location data (geospatial and audio data). 44,51,55 Samples vary for each study depending on their purpose.…”

“…Other outcomes included medication adherence, 49,53,58 fall prevention, 42 30-day morbidity, 51,52,59 service utilization, 55 and risk prediction. 41,48,49,52,59 Variables to assess these outcomes included patient demographics, [48][49][50]52,54 use of medications, 41,53 clinical diagnoses, 45,46,50,53,54,56,59 distance from health services, 45 lab results, 54 modifiable risk factors, 46,50,51 images, 42,43 self-reported outcomes, 46,48 sound detection, 44 and various flags in unstructured data. 45,47,[56][57][58] Studies used various data science methods to evaluate their outcomes.…”