External validation and update of a prognostic model to predict mortality in hospitalized adults with RSV: A retrospective Dutch cohort study

Abstract Respiratory syncytial virus (RSV) causes significant mortality in hospitalized adults. Prediction of poor outcomes improves targeted management and clinical outcomes. We externally validated and updated existing models to predict poor outcome in hospitalized RSV‐infected adults. In this single center, retrospective, observational cohort study, we included hospitalized adults with respiratory tract infections (RTIs) and a positive polymerase chain reaction for RSV (A/B) on respiratory tract samples (2005‐2018). We validated existing prediction models and updated the best discriminating model by revision, recalibration, and incremental value testing. We included 192 RSV‐infected patients (median age 60.7 years, 57% male, 65% immunocompromised, and 43% with lower RTI). Sixteen patients (8%) died within 30 days. During hospitalization, 16 (8%) died, 30 (16%) were admitted to intensive care unit, 21 (11%) needed invasive mechanical ventilation, and 5 (3%) noninvasive positive pressure ventilation. Existing models performed moderately at external validation, with C‐statistics 0.6 to 0.7 and moderate calibration. Updating to a model including lower RTI, chronic pulmonary disease, temperature, confusion and urea, increased the C‐statistic to 0.76 (95% confidence interval, 0.61‐0.91) to predict in‐hospital mortality. In conclusion, existing models to predict poor prognosis among hospitalized RSV‐infected adults perform moderately at external validation. A prognostic model may help to identify and treat RSV‐infected adults at high‐risk of death.


| INTRODUCTION
There is increasing evidence that respiratory syncytial virus (RSV) is a common cause of respiratory tract infections (RTI) in adult patients, 1 often with a complicated course of disease. [2][3][4][5] Among hospitalized elderly ≥65 years of age mortality is as high as 8%, 2 but among high-risk groups as patients with chronic heart or lung disease, long-term care facility residents and immunocompromised patients as lung or hematopoietic cell transplant (HCT) recipients, RSV may even lead to mortality rates over 50%. 2 With the widespread implementation of rapid tests for respiratory viruses in-hospital care settings, early detection of RSV enables early treatment with either aerosolized or oral ribavirin 6,[8][9][10][11][12][13][14] and future medicaments as fusion protein inhibitors (eg, presatovir), nucleoside inhibitors (eg, lumicitabine), 15 and viral replication lowering immunoglobulins (eg, palivizumab), which might have an additional positive effect to ribavirin. 11,[16][17][18] Ideally, in light of effectivity and potential side effects, treatment should be targeted to patients at the highest risk of a lifethreatening infection. Identification of RSV-infected patients at high-risk of death is therefore necessary to improve targeted therapy and clinical outcomes. In addition, the prediction of individual prognosis improves decision making on the necessity to apply supportive in-hospital management as intensive care unit (ICU) admission and strict isolation procedures. 3 However, a validated prognostic model to identify adult patients with a high mortality risk is not available. Therefore, we aimed to establish factors associated with poor prognosis and externally validate and update existing models to predict mortality in hospitalized RSV-infected adults.

| Study population
We performed a single center cohort study to validate prognostic models for poor outcomes in hospitalized adults with RSV. In the validation cohort we included adult patients (≥18 years) with a laboratory confirmed community acquired RSV-infection between  (Table S1). 23

| External validation
We searched available literature on predictive models for RSV prognosis in the MEDLINE. We aimed to validate models predicting mortality, but also included studies using a composite outcome including mortality. For the external validation, we applied the included original prognostic models to our study cohort exactly as they were published, with similar definitions of predictor variables and outcomes (Table S2). [24][25][26][27] If the intercept from the original model was not reported, we calculated a new intercept by recalibration. We compared the discriminative ability of the models using the Harrell's C-statistic. Calibration of the models was assessed in calibration plots. [28][29][30]

| Model update
We selected the model with the best discrimination and calibration for further updating. 24 In view of increasingly shorter turnaround times of molecular diagnostics and increased effectiveness of antiviral treatment when given at an early stage, 6 we first removed any eventual predictors that could not be assessed at the time of presentation/RSV diagnosis, eg, bacterial coinfection. Furthermore, we replaced binary predictors with continuous to avoid loss of information, eg, temperature instead of fever. Next, we recalibrated the calibration slope and intercept by refitting this adapted model in the validation cohort. Consequently, we tested the incremental value of the model by adding objectively assessable predefined predictor variables (age, gender, urea, confusion, cardiovascular comorbidities, immunocompromised status, and the number of other comorbidities), based on the existing prognostic models for poor outcomes in patients with positive influenza virus. 26,27,31 We performed backward variable selection based on the Akaike information criterion and Occam's razor principle. Finally, we performed internal validation with optimism correction by bootstrap. 32 Discrimination and calibration of this final updated and extended model was assessed for inhospital mortality, 30-day mortality and a composite outcome consisting of in-hospital death, ICU-admission and/or need for mechanical ventilation separately. Furthermore, we performed a decision curve analysis to provide insight into the range of predicted risks for which the final model results in better clinical decision making, eg, is better than either classifying all or none of the patients as having the outcome. 33
Demographics and characteristics of the included patients are displayed in Table 1

| External validation
We found five studies that developed a prognostic model for hospitalized RSV-infected adult patients, of which two to predict mortality 24,25 and three to predict disease progression to a lower RTI 34-37 ( Figure S2). The two models to predict mortality were included for external validation. A detailed overview of these two   Figure 2B).

| Model update
We updated and extended the model of Park et al, 24    To our knowledge, this is one of the largest studies in RSVinfected adult patients in a hospital care setting. We found a high percentage of 8% in-hospital mortality, which is in line with 8% to 9% mortality rates reported in former publications. 2,6,24,25 This high mortality rate underlines the great importance of targeted treatment for these patients. Also, this is the first study to externally validate existing models to predict poor prognosis in RSV-infected hospitalized adult patients, and allows for a head-to-head comparison of two published models. Unfortunately, model performance in the development cohorts was not described, 30