Levodopa is a time-sensitive therapy for people with Parkinson disease and maintaining scheduled dosing in the hospital is clinically important. The relationship between care setting—specifically intensive care unit (ICU) exposure—and reliable levodopa administration is uncertain. The authors aimed to quantify levodopa dosing fidelity in hospitalized patients, test the association of ICU exposure with delayed or omitted doses using clustering-aware methods, and evaluate whether documented enteral or parenteral access type is associated with dosing reliability.
This retrospective cohort analysis used the MIMIC-IV database covering 2011–2022. Inclusion criteria were adult patients with Parkinson disease who had at least one scheduled levodopa dose during hospitalization. The final cohort comprised 1,665 admissions contributing 39,322 scheduled levodopa doses.
The primary on-time dosing definition used a 60-minute window around the scheduled time. A secondary, stricter symmetric definition used ±15 minutes and produced a different classification of mistimed doses. Outcomes focused on delayed or omitted levodopa doses relative to scheduled times.
To address within-patient correlation of repeated dosing observations, the authors used patient-clustered generalized estimating equation (GEE) models when testing the association between ICU exposure and dosing reliability. Among ICU-exposed doses, access type categories included normal (oral), tube feeding, parenteral nutrition, and NPO. A fully adjusted model evaluated access-type associations. Specificity was tested by comparing results for levodopa with an active-comparator medication (statins) restricted to the ICU. Additional sensitivity checks and route/formulation re-analyses were performed; where sample sizes were small, results were reported as limited or inconclusive.
Across 39,322 scheduled levodopa doses, 79.8% met the primary 60-minute on-time definition. When applying the stricter ±15-minute symmetric definition, a larger proportion of doses were classified as mistimed—68.8% by that tighter standard.
In unclustered comparisons, ICU exposure might appear associated with dosing unreliability; however, after accounting for patient-level clustering with the GEE approach, ICU admission alone was not associated with delayed or omitted dosing. The patient-clustered odds ratio was 0.87 with a 95% confidence interval of 0.74–1.01, indicating no statistically robust association after clustering.
When analysis was restricted to doses administered during ICU exposure and adjusted for relevant covariates, documented access type showed differential associations with dosing reliability:
NPO (nothing by mouth) status was associated with higher odds of delayed or omitted levodopa dosing (adjusted OR 1.89; 95% CI, 1.36–2.62).
Tube feeding was associated with lower odds of delayed or omitted dosing (adjusted OR 0.62; 95% CI, 0.42–0.92).
Parenteral nutrition and the pooled category of access types were evaluated, and the authors emphasize that access type should not be treated as a single homogeneous category when assessing dosing reliability in the ICU.
To test specificity of the access-type associations, the authors restricted a comparator analysis to the ICU and used statins as an active-comparator medication. The comparator produced a directionally consistent interaction but was described as inconclusive, limited by the available sample (reported as 92 patients in the comparator check). The authors note that this specificity check was valid only within the ICU and that results did not definitively attribute the observed associations uniquely to levodopa.
The investigators performed route-order and formulation-confound checks. Among immediate-release formulations, a route-order association was not observed; this analysis was limited by very small sample sizes in some subgroups (for example, an OR reported for 4 patients in one route-order re-analysis). The report states that some sensitivity and subgroup checks were underpowered and therefore inconclusive.
In this large MIMIC-IV cohort, ICU admission by itself did not explain levodopa dosing unreliability after accounting for within-patient clustering. Instead, among ICU-exposed doses, documented access type—notably NPO and tube feeding—showed associations with delayed or omitted dosing: NPO with higher odds of missed/delayed doses and tube feeding with lower odds. The authors caution that comparator- medication specificity checks were directionally consistent but inconclusive, and that several subgroup analyses were limited by small samples. They also note that the work is a preprint and not peer reviewed; findings should not directly alter clinical practice without further confirmation.
MIMIC-IV v3.1 is available to credentialed researchers through PhysioNet after completion of required training and data use agreements. The authors have made analysis code available on GitHub; the repository does not contain patient-level data, consistent with the MIMIC-IV data use agreement. The MIMIC-IV dataset was approved by the institutional review boards of Beth Israel Deaconess Medical Center and MIT with a waiver of individual consent because data are de-identified. The authors declared no competing interests.