Characteristics of COVID-19 patients admitted to a tertiary care hospital in Pune, India, and cost-effective predictors of intensive care treatment requirement

BackgroundMaharashtra is one of the worst affected states in this pandemic.2 As of 30th September, Maharashtra has in total 1.4 million cases with 38,000 deaths. Objective was to study associations of severity of disease and need for ICU treatment in COVID-19 patients. MethodsA retrospective study of clinical course in 800 hospitalized COVID-19 patients, and a predictive model of need for ICU treatment. Eight hundred consecutive patients admitted with confirmed COVID-19 disease. ResultsAverage age was 41 years, 16% were <20 years of age, 55% were male, 50% were asymptomatic and 16% had at least one comorbidity. Using MoHFW India severity guidelines, 73% patients had mild, 6% moderate and 20% severe disease. Severity was associated with higher age, symptomatic presentation, elevated neutrophil and reduced lymphocyte counts and elevated inflammatory markers. Seventy-seven patients needed ICU treatment: they were older (56 years), more symptomatic and had lower SpO2 and abnormal chest X-ray and deranged hematology and biochemistry at admission. A model trained on the first 500 patients, using above variables predicted need for ICU treatment with sensitivity 80%, specificity 88% in subsequent 300 patients; exclusion of expensive laboratory tests did not affect accuracy. ConclusionIn the early phase of COVID- 19 epidemic, a significant proportion of hospitalized patients were young and asymptomatic. Need for ICU treatment was predicted by simple measures including higher age, symptomatic onset, low SpO2 and abnormal chest X-ray. We propose a cost-effective model for referring patients for treatment at specialized COVID-19 hospitals. Key MessagesO_LIOf 800 patients, 73% had mild, 6% moderate and 20% had severe disease. C_LIO_LISeventy-seven patients (9.6%) required ICU treatment, 25 (3%) died. C_LIO_LIICU treatment was predicted by higher age, more symptomatic presentation, lower SpO2 and pneumonia on chest X-ray at admission. C_LIO_LIA machine learning model features in first 500 patients accurately predicted ICU treatment in subsequent 300 patients. C_LIO_LIA good clinical protocol, SpO2 and chest X-ray are adequate to predict and triage COVID-19 patients for hospital admissions in resource poor environments. C_LI