Prediction of SARS-CoV-2-positivity from million-scale complete blood counts using machine learning

Abstract

Background The Complete Blood Count (CBC) is a commonly used low-cost test that measures white blood cells, red blood cells, and platelets in a person’s blood. It is a useful tool to support medical decisions, as intrinsic variations of each analyte bring relevant insights regarding potential diseases. In this study, we aimed at developing machine learning models for COVID-19 diagnosis through CBCs, unlocking the predictive power of non-linear relationships between multiple blood analytes. Methods We collected 809,254 CBCs and 1,088,385 RT-PCR tests for SARS-Cov-2, of which 21% (234,466) were positive, from 900,220 unique individuals. To properly screen COVID-19, we also collected 120,807 CBCs of 16,940 individuals who tested positive for other respiratory viruses. We proposed an ensemble procedure that combines machine learning models for different respiratory infections and analyzed the results in both the first and second waves of COVID-19 cases in Brazil. Results We obtain a high-performance AUROC of 90 + % for validations in both scenarios. We show that models built solely of SARS-Cov-2 data are biased, performing poorly in the presence of infections due to other RNA respiratory viruses. Conclusions We demonstrate the potential of a novel machine learning approach for COVID-19 diagnosis based on a CBC and show that aggregating information about other respiratory diseases was essential to guarantee robustness in the results. Given its versatile nature, low cost, and speed, we believe that our tool can be particularly useful in a variety of scenarios—both during the pandemic and after.

Ensemble Learning through Rashomon Sets

“…We also demonstrate the necessity of maintaining a model as up-to-date as possible to allow any machine learning model to keep up with the different stages of a pandemic surge. Our model retains high-performance values across multiple evaluation scenarios and on simulations with varying prevalences of COVID-19, properly differentiating Sars-CoV-2 from other confounding viruses, thus demonstrating the robustness of our approach presented in our work [Zuin et al, 2022a].…”

“…These properties imply a partial ordering of the graph starting from the root node, which allows us to search it in an orderly manner. It has been shown that this modeling approach is effective for the task at hand [Zuin et al, 2021[Zuin et al, , 2022a. We can, for example, apply the A* algorithm [Hart et al, 1968] employing as heuristic the performance of the model represented by the feature set of a given vertex and the Jensen-Shannon distance to the predictions of the remaining Rashamon subgroups clusteroids.…”

Unlocking the complete blood count as a risk stratification tool for breast cancer using machine learning: a large scale retrospective study

“…This process of enumerating models and selecting the best-performing one based on AUC continues, with one additional feature being incorporated at each step. The procedure halts once the improvement in AUC no longer exceeds its standard deviation, ensuring a balance between model complexity and performance 14 , 15 .…”

Cardiovascular and Renal Comorbidities Included into Neural Networks Predict the Outcome in COVID-19 Patients Admitted to an Intensive Care Unit: Three-Center, Cross-Validation, Age- and Sex-Matched Study

“…In contrast, complete blood count and routine biochemical markers have been less represented in ML approaches, though reaching similar AUROC values in combination with clinical risk factors such as comorbidities [ 8 , 9 , 10 , 11 , 12 ]. Moreover, an ensemble of laboratory markers has been suggested even for COVID-19 diagnosis [ 13 , 14 , 15 , 16 ], but the efficacy of this approach seems limited to pandemics, as having an unacceptably high risk of false positives related to other infectious diseases [ 17 ]. However, the rapid prognosis of COVID-19 using routine markers such as complete blood count or C-reactive protein might be a cost-efficient solution for developing and least-developed countries in case of pandemics caused by other SARS-CoV-2 mutants.…”