In this study, we created a predictive model for the risk of death in patients with ILD receiving long-term glucocorticoid therapy within 30 days of contracting pneumonia. The best features were determined by using Lasso regression analysis to determine the use of ventilators, respiratory failure, vasoactive drug use, and lymphocytopenia. In both the training set (AUC = 0.897) and the validation set (AUC = 0.903), the model showed good predictive ability, indicating its possible clinical utility.
ILD is a collection of diverse lung conditions marked by inflammation and, occasionally, fibrosis [20]. These conditions cause hypoxemia, which can lead to respiratory failure, dyspnea, coughing, abnormal gas exchange, and restrictive physiology, which manifests as decreased lung capacity. Genetic variations, environmental exposures, and systemic diseases are some of the factors that contribute to the pathogenesis of ILD [21, 22]. The incidence of ILD varies greatly between global regions and is trending gradually higher [23, 24].
In the treatment of ILD, glucocorticoids are a crucial therapeutic choice. Although their effectiveness varies depending on the type of ILD, glucocorticoids typically help patients with lung function, clinical symptoms, and inflammatory response reduction. It is commonly acknowledged that one of the first-line therapies for ILD is glucocorticoids [25]. They work by reducing inflammatory reactions and over-stimulating the immune system, which lessens lung damage and enhances the quality of life for patients [26]. The use of glucocorticoids is not risk-free, though. Extended use of glucocorticoids can lead to immunosuppression in patients and a faster course of disease because it can decrease drug sensitivity, raise the risk of infections, and possibly affect the immune system [27]. Predicting death and prognosis in these patients is therefore especially important.
Both congenital and secondary lymphocytopenia can occur; secondary lymphocytopenia frequently results from long-term glucocorticoid use or from recurrent infections [28]. Individuals with immune-mediated lymphocytopenia who receive long-term glucocorticoid therapy for their underlying illness are also more susceptible to lymphocytopenia [29]. As essential immune cells involved in recognizing and getting rid of pathogens like bacteria and viruses, lymphocytes are essential to the immune system [30]. A common symptom of acute respiratory distress syndrome (ARDS), which is frequently linked to severe pneumonia, is respiratory failure. Research has also indicated that the severity of respiratory failure is associated with patient mortality [31, 32]. Respiratory failure is also a common complication in the terminal stage of ILD patients with pneumonia, who may experience exacerbated dyspnea and hypoxemia, necessitating the use of a ventilator to maintain oxygenation and assist ventilation. The presence of respiratory failure and ventilator use may indicate a relatively severe state of ILD or a more severe degree of infection, both of which can raise the risk of death [33]. When severe pneumonia results in septic shock, vasoactive drugs are essential in maintaining basic vital signs in the respiratory, circulatory, and metabolic systems [34]. When pneumonia patients develop hemodynamic instability and shock, the probability of progressing to severe pneumonia increases significantly, further elevating the risk of death [35]. Therefore, these four variables were used as predictors to construct the predictive model in this study. The decision curve analysis showed that if the threshold probability for an individual is less than 77%, using the model developed in this study provides more benefits than either treating all or not treating.
Numerous mortality prediction models for pneumonia, particularly those pertaining to COVID-19, have been published thus far. However, most of them are predictions for general patients, and there are few prediction models for patients with ILD to predict pneumonia. Moreover, the majority of these models employ a substantial number of predictors. For instance, Zhao et al.'s study utilized seven variables—including heart failure, procalcitonin, lactate dehydrogenase, chronic obstructive pulmonary disease, pulse oxygen saturation, heart rate, and age—as predictors for the mortality risk of COVID-19 patients, encompassing two laboratory tests, two vital signs, two diagnoses, and one patient demographic[36]. In contrast, the prediction model presented in this paper necessitates only four predictors, of which only respiratory failure necessitates laboratory testing. The remaining three predictors are rooted in patients' medical histories and objective factors related to their treatment protocols, offering a more expedient and convenient approach that does not oblige the delay of awaiting test results for prediction.
It is important to recognize the limitations of this study. Firstly, not all patients in the raw data underwent a full set of pathogen tests, suggesting that the detection and diagnosis of pathogens in patients may be incomplete. Secondly, the detection of pathogens was not completed immediately upon admission, and some pathogens may only be detected several days after admission, thereby increasing the risk of nosocomial infections. Moreover, the study was validated using an internal validation set, and the consequences have not been validated with external cases, so caution should be exercised in its clinical application. Additionally, the database was gathered from hospitals in China, and its applicability to other countries requires further validation. But our research findings align with clinical observations and the currently published literature, offering valuable insights into the prevention of mortality and prognosis for patients with ILD who develop pneumonia while undergoing long-term glucocorticoid therapy.