The review of literature showed that no research has been done so far on the predictive determinants of overall survival among re-infected COVID-19 patients. Only the systematic review conducted by SeyedAlinaghi et al. was a comprehensive study which assessed the risk of COVID-19 re-infection (6). They found thirty-one eligible studies of which eight studies described the patients who recovered from COVID-19 re-infection and only one study reported death among them. However, the majority of the published works (26 studies) did not present any extra information about the patients’ status (i.e. death or discharge) (6).
The underlying diseases, clinical conditions, use of glucocorticoids, and secondary bacterial infection were identified as the independent risk factors of COVID-19 re-infection (6, 26, 27). Interestingly, in the COVID-19 re-infection dataset of the current study, about 70% of the cases used steroids including dexamethasone, hydrocortisone, and methylprednisolone as adjuvant therapy. This confirms that using steroids and glucocorticoids increases the probability of re-infection among COVID-19 patients. Moreover, although re-infection is possible, it should be noted that the re-infection or reactivation diagnosed in some patients might in fact be a false negative at the time of discharge or not meeting the discharge criteria completely. On the other hand, three main reasons including short-lived, ineffective, and strain-specific immune responses may lead to a positive PCR test result (28, 29).
Recent studies have reported that some patients who had recovered from COVID-19 had a positive PCR test result for the second time (5, 8, 30–36). For instance, it was stated in a report that 116 patients in South Korea who had recovered from COVID-19 had positive PCR test results again (33). In addition, most previously published works which described patients with COVID-19 re-infection were in the format of case reports (5, 8, 30–32, 34–36) and no studies evaluated the OS and its related predictors among these patients.
Regularization algorithms such as elastic-net and LASSO can be used to perform feature selection and to improve the prediction accuracy by shrinking the coefficients towards zero (24). In this study, two ML algorithms (elastic-net regularized Cox-adjusted PH model and backward stepwise elimination) were applied to the dataset of re-infected COVID-19 patients to predict the OS and the associated factors among them. The current study is unique in that it incorporates all regularized algorithms under the elastic-net umbrella. These algorithms created two models. One of them maximized parsimony and the other optimized the predictive power. The elastic-net Cox-adjusted PH regression kept 8 out of 35 candidate features of time to discharge or in-hospital death. The strongest predictors (i.e. the features with the highest magnitude of the estimated coefficients) included the type of patient transfer (using the EMS or not), SpO2, intubation, and triage level (level 1 vs. others). The backward elimination method further reduced the regularized model to retain four features: type of patient transfer, SpO2, WBC count, and serum creatinine.
Since no similar studies were found about the survival of re-infected COVID-19 patients, the results of this study were compared with those of the studies related to survival and the related risk factors in patients with COVID-19. The results of the current research showed that the empirical in-hospital mortality rate was 9.5%. Furthermore, the OS rates for days 7, 14, and 21 were obtained as 87.5%, 78.3%, and 52.2%, respectively, in the re-infected COVID-19 inpatients. These rates have been reported differently for COVID-19 patients in other studies (12, 15, 37). For example, Murillo-Zamora and Hernandez-Suarez found that 7-, 15-, 21- and 30-day OS rates were respectively 72.2%, 47.6%, 35.0%, and 23.9% which were lower than the results obtained in the current study (37). In another study by Sousa et al., the 24-day OS rate in 2070 patients with COVID-19 was calculated as 87.7% (15).
Regarding the laboratory findings at the time of admission, it was found that increased serum creatinine (more than 1.6 mg/dL) and increased WBC count (more than 8.5 (×109 cells/L)) were associated with a higher mortality rate in re-infected COVID-19 patients. As compared with the surviving re-infected COVID-19 patients, the levels of creatinine were independent predictors of abnormal kidney function at the time of admission in the non-surviving re-infected COVID-19 patients. The higher in-hospital mortality rate was related to the higher concentration levels of creatinine (> 1.6 mg/dL) in the patients, suggesting a worse renal function at the time of hospital admission. This finding is in line with previous studies which revealed that the concentration levels of creatinine were significantly higher among the COVID-19 patients who died (38–40).
Moradi et al. assessed the risk of one-month mortality from COVID-19 since the time of admission. They found that increased NLR and increased WBC count were associated with a higher one-month death rate. Moreover, although hypoxemia (SpO2 < 90%) increased the one-month mortality rate, this association was not significant (18). After adjustment for confounders, the results of the present study demonstrated that higher SpO2 levels (greater than 85%) after oxygen supplementation were associated with reduced mortality. In fact, profound hypoxemia (SpO2 ≤ 85%) could have a harmful effect on the OS of re-infected COVID-19 patients, increasing the risk of mortality eight-fold. The findings of the present study were consistent with previous studies in which profound hypoxemia was associated with a higher in-hospital death rate (41, 42).
Another survey by Yan et al. applied an ML-based algorithm to predict OS among 404 patients with severe COVID-19. They reported three biomarkers including lymphocyte, lactic dehydrogenase (LDH), and high-sensitivity C-reactive protein (hs-CRP) as the survival predictors with the accuracy of more than 90%. In particular, it was revealed that high levels of LDH might have an independent harmful effect on the OS rate (43).
We could not compare our results with other studies because we did not find any studies reporting transfer by EMS as an OS predictive factor. However, it could be said that the patients who were transferred to EMDs by EMS had a more severe status, increasing their mortality rate almost four-fold.
This study had several limitations which should be mentioned. We could not find any similar study in the literature to compare our findings with. Therefore, we had to compare our results with studies which used general COVID-19 datasets for their analyses. The impossibility of examining the risk factors associated with re-infection as well as the difficulty of confirming the diagnosis of COVID-19 re-infection were two other limitations of the present study. Another limitation of this study was that it was conducted during the peak period of infection especially when the virus had an active transmission chain among the populations. Hence, our findings may vary in non-pandemic conditions.