Walidacja algorytmów identificzujących epilepsję oporną na leki u weteranów USA w różnych placówkach medycznych
PubMedEpilepsia
Validation of drug-resistant epilepsy identification algorithms in US Veterans across diverse care settings
W skrócie
Badanie sprawdziło, czy komputerowe algorytmy mogą prawidłowo rozpoznać epilepsję oporną na leki (czyli taką, która nie odpowiada na standardowe leczenie) u ponad 970 weteranów wojskowych. Naukowcy testowali różne kombinacje warunków, takie jak liczba zażywanych leków, czas ich stosowania i wyniki badań (EEG, MRI), aby znaleźć najlepszy sposób na identyfikowanie pacjentów z oporną epilepsją. Rezultaty pokazały, że najlepszy algorytm potrafi prawidłowo rozpoznać oporne epilepsję w około 70% przypadków, niezależnie od tego, czy pacjent jest leczony w specjalistycznym ośrodku epilepsji czy zwykłym szpitalu.
Oryginalny abstract (angielski)
OBJECTIVE: Early recognition of drug-resistant epilepsy (DRE) is critical for timely referral to comprehensive epilepsy centers (CECs). Our prior pilot study in the Veterans Health Administration (VHA) demonstrated the feasibility of an algorithm-based approach to identify DRE within selected CECs. In this study, we validated and refined these algorithms in a broader, national cohort, including CEC and non-CEC care, to optimize algorithms for different care settings. METHODS: We conducted a multicenter, retrospective study of randomly sampled veterans with epilepsy on ≥2 antiseizure medications (ASMs) from 1999 to 2024. Chart review by 26 epilepsy providers rated DRE status. Ninety-six candidate algorithms using combinations of six criteria-number of ASMs, treatment duration, dose thresholds, pregabalin exclusion, intractable International Classification of Diseases (ICD) codes, and electroencephalography (EEG)/magnetic resonance imaging (MRI) utilization-were evaluated. Performance for DRE classification was evaluated among all veterans, and separately for CEC or non-CEC care. RESULTS: Among 970 veterans on ≥2 lifetime ASMs, 518 (53.4%) were rated as DRE, 363 (37.4%) as non-DRE, and 89 (9.2%) as undetermined. Veterans with DRE were younger (mean age = 68.9 vs. 72.0 years) and more frequently had at least one intractable ICD code (46.1% vs. 18.2%), EEG (77.8% vs. 66.7%), or MRI (78.0% vs. 65.8%) compared with non-DRE (all p-values <.001). Among all veterans, the best overall algorithm (area under the curve [AUC] = .701) identified DRE as either (1) exposure to 3+ ASMs or (2) exposure to two ASMs for at least 1 year, with the requirement of past EEG/MRI and intractable ICD code documentation. This algorithm was also highly effective among veterans who received CEC (AUC = .756) or only non-CEC care (AUC = .678). SIGNIFICANCE: This national validation study confirms that medication-based algorithms can identify DRE within VHA administrative data across care settings (CECs and non-CECs). This supports the implementation of a standardized DRE definition for quality improvement, surveillance, and referral optimization within the VHA.