Machine learning uncovers blood test patterns subphenotypes at hospital admission discerning increased 30-day ICU mortality rates in COVID-19 elderly patients

  1. Lexin Zhou
  2. Nekane Romero-García
  3. Rafael Badenes
  4. Teresa García Morales
  5. David Lora
  6. Agustín Gómez de la Cámara
  7. Francisco T García Ruiz
  8. Juan M García-Gómez
  9. Carlos Sáez

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Elderly patients with COVID-19 are among the most numerous populations being admitted in the ICU due to its high mortality rate and high comorbidity incidence. An early severity risk stratification at hospital admission could help optimize ICU usage towards those more vulnerable and critically ill patients.

Methods

Of 503 Spanish patients aged>64 years admitted in the ICU between 26 Feb and 02 Nov 2020 in two Spanish hospitals, we included 193 quality-controlled patients. The subphenotyping combined PCA and t-SNE dimensionality reduction methods to maximize non-linear correlation and reduce noise among age and full blood count tests (FBC) at hospital admission, followed by hierarchical clustering.

Findings

We identified five subphenotypes (Eld-ICU-COV19 clusters) with heterogeneous FBC patterns associated to significantly disparate 30-day ICU mortality rates ranging from 2% in a healthy cluster to 44% in a severe cluster, along three moderate clusters.

Interpretations

To our knowledge, this is the first study using age and FBC at hospital admission to early stratify the risk of death in ICU at 30 days in elderly patients. Our results provide guidance to comprehend the phenotypic classification and disparate severity patterns among elderly ICU patients with COVID-19, based only on age and FBC, that have the potential to establish target groups for early risk stratification or early triage systems to provide personalized treatments or aid the decision-making during resource allocation process for each target Eld-ICU-COV19 cluster, especially in those circumstances with resource scarcity problem.

Funding

FONDO SUPERA COVID-19 by CRUE-Santander Bank grant SUBCOVERWD-19.

Research in context

Evidence before this study

We searched on PubMed and Google Scholar using the search terms “COVID-19”, “SARS-CoV2”, “phenotypes” for research published between 2020 to 2022, with no language restriction, to detect any published study identifying and characterizing phenotypes among ICU COVID-19 patients. A previous COVID-19 phenotyping study found three phenotypes from hospitalized patients associated with significantly disparate 30-day mortality rates (ranging from 2·5 to 60·7%). However, it seems to become harder to find phenotypes with discriminative mortality rates among ICU COVID-19 patients. For example, we found one study that uncovered two phenotypes from 39 ICU COVID-19 patients based on biomarkers with 39% and 63% mortality rates, but such difference was not statistically significant. We also found another study with more success that uncovered two ICU COVID-19 phenotypes using two different trajectories with somehow disparate 28-day mortality rates of 27% versus 37% (Ventilatory ratio trajectories) and of 25% versus 39% (mechanical power trajectories).

Added value of this study

To our knowledge, this is the first study that uses age and laboratory results at hospital admission (i.e., before ICU admission) in elderly patients to early stratify, prior ICU admission, the risk of death in ICU at 30 days. We classified 193 patients with COVID-19, based on age and ten Full Blood Count (FBC) tests, into five subphenotypes (one healthy, three moderate, and one severe) that showed significantly disparate 30-day ICU mortality rates from 2% to 44%.

Implications of all the available evidence

Identifying, from elderly ICU patients with COVID-19 (Eld-ICU-COV19), subphenotypes could spur further investigation to analyze the potential differences in their underlying disease mechanisms, acquire better phenotypical understanding among Eld-ICU-COV19 toward better decision-making in distributing the limited resources (including both logistic and medical) as well as shedding light on tailoring personalized treatment for each specific target subgroup in future medical research and clinical trial.

Article activity feed

  1. Version published to 10.1101/2022.05.10.22274889v2 on medRxiv
  2. ScreenIT

    SciScore for 10.1101/2022.05.10.22274889: (What is this?)

    Please note, not all rigor criteria are appropriate for all manuscripts.

    Table 1: Rigor

    EthicsIRB: The use of data was approved by the Ethical Committees of the two hospitals and the Universitat Politècnica de València.
    Sex as a biological variablenot detected.
    Randomizationnot detected.
    Blindingnot detected.
    Power Analysisnot detected.

    Table 2: Resources

    No key resources detected.

    Results from OddPub: We did not detect open data. We also did not detect open code. Researchers are encouraged to share open data when possible (see Nature blog).

    Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:
    One limitation of this study is its sample size. The small number of patients in some subphenotypes may make the comparative statistics sometimes difficult to interpret (e.g., cluster 4 only has 14 patients). In addition, our data are from two hospitals, which favors generalization, although future studies from third hospitals may benefit as external validations. In summary, by using Machine Learning we identified five Eld-ICU-COV19 subphenotypes with discriminative FBC patterns alongside age. Of which, we found one healthy cluster –where nearly all patients survived within 30-day after ICU admission, one severe cluster –where nearly half of the patients lost their life within 30-day after ICU admission– and three moderate clusters –whose 30-day ICU mortality is similar to the populational level– with one that has the potential of being categorized into the severe category. Our results can provide guidance to comprehend the phenotypic classification and disparate severity patterns among elderly ICU patients with COVID-19, based only on age and FBC tests, that have the potential to establish target groups for an early risk stratification prior to ICU admission or early triage systems to provide personalized treatments or aid the decision-making during resource allocation process for each target Eld-ICU-COV19 group, especially in those circumstances with resource scarcity problem.

    Results from TrialIdentifier: No clinical trial numbers were referenced.

    Results from Barzooka: We did not find any issues relating to the usage of bar graphs.

    Results from JetFighter: We did not find any issues relating to colormaps.

    Results from rtransparent:
    • Thank you for including a conflict of interest statement. Authors are encouraged to include this statement when submitting to a journal.
    • Thank you for including a funding statement. Authors are encouraged to include this statement when submitting to a journal.
    • No protocol registration statement was detected.

    Results from scite Reference Check: We found no unreliable references.

    About SciScore

    SciScore is an automated tool that is designed to assist expert reviewers by finding and presenting formulaic information scattered throughout a paper in a standard, easy to digest format. SciScore checks for the presence and correctness of RRIDs (research resource identifiers), and for rigor criteria such as sex and investigator blinding. For details on the theoretical underpinning of rigor criteria and the tools shown here, including references cited, please follow this link.

  3. Version published to 10.1101/2022.05.10.22274889v1 on medRxiv