COVID-19 disease—Temporal analyses of complete blood count parameters over course of illness, and relationship to patient demographics and management outcomes in survivors and non-survivors: A longitudinal descriptive cohort study

Background Detailed temporal analyses of complete (full) blood count (CBC) parameters, their evolution and relationship to patient age, gender, co-morbidities and management outcomes in survivors and non-survivors with COVID-19 disease, could identify prognostic clinical biomarkers. Methods From 29 January 2020 until 28 March 2020, we performed a longitudinal cohort study of COVID-19 inpatients at the Italian National Institute for Infectious Diseases, Rome, Italy. 9 CBC parameters were studied as continuous variables [neutrophils, lymphocytes, monocytes, platelets, mean platelet volume, red blood cell count, haemoglobin concentration, mean red blood cell volume and red blood cell distribution width (RDW %)]. Model-based punctual estimates, as average of all patients’ values, and differences between survivors and non-survivors, overall, and by co-morbidities, at specific times after symptoms, with relative 95% CI and P-values, were obtained by marginal prediction and ANOVA- style joint tests. All analyses were carried out by STATA 15 statistical package. Main findings 379 COVID-19 patients [273 (72% were male; mean age was 61.67 (SD 15.60)] were enrolled and 1,805 measures per parameter were analysed. Neutrophils’ counts were on average significantly higher in non-survivors than in survivors (P<0.001) and lymphocytes were on average higher in survivors (P<0.001). These differences were time dependent. Average platelets’ counts (P<0.001) and median platelets’ volume (P<0.001) were significantly different in survivors and non-survivors. The differences were time dependent and consistent with acute inflammation followed either by recovery or by death. Anaemia with anisocytosis was observed in the later phase of COVID-19 disease in non-survivors only. Mortality was significantly higher in patients with diabetes (OR = 3.28; 95%CI 1.51–7.13; p = 0.005), obesity (OR = 3.89; 95%CI 1.51–10.04; p = 0.010), chronic renal failure (OR = 9.23; 95%CI 3.49–24.36; p = 0.001), COPD (OR = 2.47; 95% IC 1.13–5.43; p = 0.033), cardiovascular diseases (OR = 4.46; 95%CI 2.25–8.86; p = 0.001), and those >60 years (OR = 4.21; 95%CI 1.82–9.77; p = 0.001). Age (OR = 2.59; 95%CI 1.04–6.45; p = 0.042), obesity (OR = 5.13; 95%CI 1.81–14.50; p = 0.002), renal chronic failure (OR = 5.20; 95%CI 1.80–14.97; p = 0.002) and cardiovascular diseases (OR 2.79; 95%CI 1.29–6.03; p = 0.009) were independently associated with poor clinical outcome at 30 days after symptoms’ onset. Interpretation Increased neutrophil counts, reduced lymphocyte counts, increased median platelet volume and anaemia with anisocytosis, are poor prognostic indicators for COVID19, after adjusting for the confounding effect of obesity, chronic renal failure, COPD, cardiovascular diseases and age >60 years.


6) Please provide figures of better quality:
We have provided all the figures (1, 2, 3, 4) in TIFF format. 7) Please define "history of neoplasm". Does it mean current chemotherapy or those receiving chemotherapy and those under surveillance are included? Do you have any data regarding the patients currently receiving chemotherapy (eg last 14 days from symptom onset) compared to the others?
As per reviewer recommendation, we have added in the manuscript the definition of "history of neoplasm" defined as "any anamnestic data of oncological disease" (current line 183).
Nineteen patients of the cohort had a positive anamnesis for oncological diseases. None of them received chemotherapy during the admission for COVID-19 and/or in the 14 days before the symptoms' onset. Out of these 19 patients, 5 died. We have added in the background a sentence supporting the rational of the study, based on the review indicated by the reviewer stating that: "As underlined by a recent review, COVID-19 has a significant impact on the hematopoietic system: lymphopenia, neutrophil/lymphocyte ratio and peak platelet/lymphocyte ratio may be considered as cardinal laboratory findings, with prognostic potential" (current lines 77 -80), and we have added the related reference (ref. 5). 11) Do you have any data regarding CRP and PCT values in different time points in order to correlate them with the corresponding neutrophil counts?

8) Please expand the
Data regarding CRP and PCT values in different time points would have definitively been extremely interesting. Unfortunately, we didn't have the possibility to analyze these data, and we have mentioned it as a limitation of the study (current lines 337-339). 12) In the limitations please make a comment regarding the 72% male percentage in the study cohort and if this has any impact on the interpretation of the results.
As suggested by the reviewer, in the discussion we have added the higher percentage of males as possible limitation of the study, explaining that "clinical outcomes can be influenced by pre-existing comorbidities, such as hypertension, cardiovascular disease and diabetes, which tended to be more frequent and more severe in men" (current lines 335 -337) and we have added a related reference 13) Several significant associations have emerged between potential confounders and CBC parameters, as described in the results. Please comment and elaborate more on potential explanations in the discussion.
We thank the reviewer for this remark which allowed us to add potential explanations regarding the association between potential confounders and CBC parameters at the end of the discussion (current lines 324 -329): "The emerged significant associations between potential confounders and CBC parameters might have several different explanations. The association between age and alteration of WBCs' and platelets' parameters, mainly lymphopenia, probably reflects the severity of COVID-19. Instead, the association between age, chronical renal failure and cardiovascular diseases with the alteration of the red blood cells' parameters, mainly anemia and anisocytosis, probably reflects the stage of the chronical conditions".
In support of these explanations we have added 3 new references: 27. Linton, P., Dorshkind, K. Age-related changes in lymphocyte development and function. Nat 1) The manuscript can be shortened substantially, especially the results and the discussion. In fact, the results are presented in a very descriptive way, and are already partly discussed.
According to the reviewer's suggestion, we have shortened results and discussion, eliminating the overlapping parts.
2) As stated in the background section, "complete blood count provides vital parameters which can inform presence of infection, response to treatment, …". In the following analysis, laboratory/clinical signs of infection and different treatment were not considered. I suggest highlighting it among the limits of the study.
As suggested by the reviewer, we have highlighted that analyzing only CBC parameters without other laboratory/clinical signs of infection, and without considering different treatment provided, represents a limitation of the study (lines 337 -339).
3) Editing for grammar and language will be helpful.
We have reviewed and corrected grammar and syntax errors (corrections marked in track changes in the manuscript).
Specific comments: 4) Study design: I am still a bit confused by the protocol -were laboratory tests performed with fixed timing (e.g. once per week)? If not, how many blood tests per patient were considered at least and at most? In fact, there is a very interesting day by day analysis (resumed on the Tables), but there is no mention of the number of tests considered for each single day. I am concerned that this might impact the strength of the results for the first days after symptom onset, when probably just a minority of patients was attended.
Thank you for this question that permitted us to better emphasize technical issues of the statistical model. As correctly suggested by the reviewer, the number of measures vary among patients. To deal with this issue we have used a mixed effect multilevel model (MEML) for unbalanced samples. This kind of models are among the most robust statistical techniques for dealing with sparse repeated measures that usually come from clinical dataset made of routine collected data. In particular our model has a random coefficient for intercept for dealing with correlation at the level of the patients; a random slope coefficient at the level of time for dealing with correlation between time and the random intercept; an unstructured covariance matrix coefficient that is specifically set for dealing with potential additional issues due to unbalanced sampling, consequent to unequal number of measures for the different subjects (i.e no assumption on variance-covariance structure was made).
We have already validated this model in previous studies that dealt with repeated spare measures: couldn't find a reference for Table 2 inside the text.
As suggested by the reviewer, we have added a legend to explain the acronyms in tables 2, 3 and 4: llb= low limit bound, ulb= lower limit bound.
The reference to table 2 is in the results section in the chapter: Leukocytes parameters kinetics (current line 192).