ABSTRACT
Patients with advanced chronic kidney disease (CKD) stage 4-5 face unknown progression rates to End Stage Renal Disease (ESRD) with elevated baseline mortality. Hemodialysis preparation requires surgical planning months in advance, and many patients may pass away before reaching ESRD. Improved understanding of survival probability in the near future could help physicians and patients in the shared decision making on the risks and benefits of dialysis vs conservative care. Patients from Kaiser Southern California Electronic Health Record (EHR) with CKD Stage 4-5 between 1/1/2010 – 12/31/2018 were selected for our initial training population. We picked an XGBoost model as it offered the best combination of accuracy and interpretability. Our features included aggregations of demographics, comorbidities calculated based on the Elixhauser comorbidity index, common labs, vitals, and past utilization data. On March 10, 2020, 16,267 current Stage 4-5 CKD patients at Kaiser Southern California were scored with the model . From March 11, 2020 to March 10, 2021, a 1-year prospective study was performed to assess the accuracy of the predictive model. At the conclusion of the 1-year observation, we assessed the model’s predictions against the actual survival data. The machine learning survival model achieved an AUC of .73 in the prospective study. We computed an optimal cut-point based on the probability prediction threshold that maximized the sum of sensitivity and specificity. At this level, the model achieved an accuracy of 70%, sensitivity of 63%, specificity of 72%, and precision of 25%, in predicting 12 month survival for individuals with advanced CKD stage 4-5. Despite unforeseen COVID-19 pandemic, our model achieved predictive accuracy for 1-year survival in CKD stage 4-5 patients prospectively. Machine learning based probabilistic forecasting can be used to better inform decisions regarding CKD management.
ABSTRACT
Background: Use of remdesivir in the treatment of dialysis patients with Coronavirus Disease 2019 (COVID-19) has been limited due to inconclusive data regarding safety outcomes among patients with severe renal impairment. For this reason, the FDA has not recommended remdesivir use in patients with eGFR < 30 ml/min per 1.73 m2. We sought to evaluate outcomes among dialysis patients with COVID-19 who received remdesivir in a real-world setting. Methods: We conducted a retrospective study of patients on hemodialysis or peritoneal dialysis hospitalized with COVID-19 between 5/1/2020 -1/31/2021 within the integrated health system of Kaiser Permanente Southern California. Patients with a COVID-19 International Classification of Diseases (ICD)-10 code: U07.1 and laboratory confirmed SARS-CoV-2 infection within 14 days prior to admission date to two days after admission date were included. The primary endpoint was 30-day all-cause mortality. Secondary endpoints were intensive care unit (ICU) stay, and evidence of acute liver injury defined as AST and/or ALT values >5x upper limit of normal. Results: A total of 486 patients (407 hemodialysis and 79 peritoneal dialysis) met inclusion criteria. Among those, 112 patients (23%) were treated with remdesivir, with median treatment time of 4 days (IQR: 2-5). Mean age was 63.8 years with 63.8% male and 63.0% Hispanic patients. There were 80.2% of patients who received treatment with steroids during hospitalization. Relative risk (RR) for all-cause 30-day mortality was 0.74 (95% CI: 0.52-1.05) in remdesivir treated patients compared to untreated patients. Acute liver injury occurred in 1.8% and 2.4% of remdesivir treated and untreated patients, respectively. ICU admissions occurred in 14.3% of remdesivir treated and 16% of untreated patients. Conclusions: Among dialysis patients hospitalized with COVID-19, treatment with remdesivir was not associated with worse outcomes in terms of liver injury or ICU admission, and demonstrated a trend (26% lower risk) toward decrease in 30-day mortality, though no statistical significance was found due to insufficient power.