Can on-admission anemia predict severe COVID-19 cases? A discussion about statistical and clinical significance

Dear Authors,

We would like to thank the authors for their commentary on our recent manuscript. We appreciate this opportunity to elaborate on some concerns raised in your letter.

We hope to clarify the first concern raised by the authors involving our sample selection method. We used “stratified random sampling,” where patients were divided into three groups based on hemoglobin cutoffs and equal number of patients were randomly sampled from each stratum. The rationale behind this was because the number of COVID-19 patients with Hb < 9 was relatively too small in the whole patient population (319/4356 to be exact). Since the research aim of our study was to evaluate the effect of on-admission anemia, as well as differing severities of anemia, on the outcomes and mortality in hospitalized COVID-19 patients, the classification of Hb was considered in the design of the study and the number of patients needed per Hb group was determined from a preliminary study of our research. Additionally, contrary to the authors' concern, random sampling in retrospective studies is neither strange nor rare [1], and is used to reduce selection bias. While we agree that including all patients in analysis is better and preferred, we were limited by the lack of resources to chart review massive data in a given study period; thus, we opted for the stratified random sampling, based on the aforementioned reasons. We hope that our response helps clarify the use of stratified random sampling in our study.

The second concern raised by the authors questions the dichotomization and categorization of hemoglobin. Again, the aim of our research was to compare Hb groups based on the World Health Organization definitions of clinical anemia vs. no anemia, and mild vs. moderate-severe anemia [2]. We only reported continuous Hb level as extra information in the descriptive demographic and clinical characteristic tables because we compared Hb groups in our main analysis. The data would have supported our findings more strongly had the continuous Hb shown significant difference between severe and no severe outcomes, but our Hb categorization is based on clinical recommendations and suits the aim of our research: to identify anemia status as a risk factor for severe outcomes.

We value your comments and agree that our paper could be further strengthened with the inclusion of the whole patient cohort. Although numerical values of hemoglobin did not appear to be significantly different among outcome groups in our research, we have demonstrated that clinical anemia and severities of anemia are risk factors to predicting unfavorable outcomes in COVID-19 patients.

Sources of support

None.

Declaration of Competing Interest

none.