Development of a prognostic model of COVID-19 severity: a population-based cohort study in Iceland

  1. Elias Eythorsson
  2. Valgerdur Bjarnadottir
  3. Hrafnhildur Linnet Runolfsdottir
  4. Dadi Helgason
  5. Ragnar Freyr Ingvarsson
  6. Helgi K. Bjornsson
  7. Lovisa Bjork Olafsdottir
  8. Solveig Bjarnadottir
  9. Arnar Snaer Agustsson
  10. Kristin Oskarsdottir
  11. Hrafn Hliddal Thorvaldsson
  12. Gudrun Kristjansdottir
  13. Aron Hjalti Bjornsson
  14. Arna R. Emilsdottir
  15. Brynja Armannsdottir
  16. Olafur Gudlaugsson
  17. Sif Hansdottir
  18. Magnus Gottfredsson
  19. Agnar Bjarnason
  20. Martin I. Sigurdsson
  21. Olafur S. Indridason
  22. Runolfur Palsson

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Background

The severity of SARS-CoV-2 infection varies from asymptomatic state to severe respiratory failure and the clinical course is difficult to predict. The aim of the study was to develop a prognostic model to predict the severity of COVID-19 in unvaccinated adults at the time of diagnosis.

Methods

All SARS-CoV-2-positive adults in Iceland were prospectively enrolled into a telehealth service at diagnosis. A multivariable proportional-odds logistic regression model was derived from information obtained during the enrollment interview of those diagnosed between February 27 and December 31, 2020 who met the inclusion criteria. Outcomes were defined on an ordinal scale: (1) no need for escalation of care during follow-up; (2) need for urgent care visit; (3) hospitalization; and (4) admission to intensive care unit (ICU) or death. Missing data were multiply imputed using chained equations and the model was internally validated using bootstrapping techniques. Decision curve analysis was performed.

Results

The prognostic model was derived from 4756 SARS-CoV-2-positive persons. In total, 375 (7.9%) only required urgent care visits, 188 (4.0%) were hospitalized and 50 (1.1%) were either admitted to ICU or died due to complications of COVID-19. The model included age, sex, body mass index (BMI), current smoking, underlying conditions, and symptoms and clinical severity score at enrollment. On internal validation, the optimism-corrected Nagelkerke’s R 2 was 23.4% (95%CI, 22.7–24.2), the C-statistic was 0.793 (95%CI, 0.789-0.797) and the calibration slope was 0.97 (95%CI, 0.96–0.98). Outcome-specific indices were for urgent care visit or worse (calibration intercept -0.04 [95%CI, -0.06 to -0.02], E max 0.014 [95%CI, 0.008–0.020]), hospitalization or worse (calibration intercept -0.06 [95%CI, -0.12 to -0.03], E max 0.018 [95%CI, 0.010–0.027]), and ICU admission or death (calibration intercept -0.10 [95%CI, -0.15 to -0.04] and E max 0.027 [95%CI, 0.013–0.041]).

Conclusion

Our prognostic model can accurately predict the later need for urgent outpatient evaluation, hospitalization, and ICU admission and death among unvaccinated SARS-CoV-2-positive adults in the general population at the time of diagnosis, using information obtained by telephone interview.

Article activity feed

  1. Version published to 10.1186/s41512-022-00130-0
  2. ScreenIT

    SciScore for 10.1101/2021.07.19.21260759: (What is this?)

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

    Table 1: Rigor

    EthicsIRB: Ethical approval: The study was approved by the National Bioethics Committee of Iceland (VSN 20-078).
    Sex as a biological variablenot detected.
    RandomizationConfidence intervals were obtained by 2000 bootstrap resamplings of a randomly selected imputed dataset.[16] Model discrimination was quantified using the C-statistic.
    Blindingnot detected.
    Power AnalysisA formal sample size calculation was not performed because all available data were used.

    Table 2: Resources

    Software and Algorithms
    SentencesResources
    The cowplot package[21] was used to create multipanel figures and tableone[22] was used to create summary tables.
    cowplot
    suggested: (cowplot, RRID:SCR_018081)

    Results from OddPub: Thank you for sharing your data.

    Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:
    Finally, CKD was associated with more severe outcomes which is consistent with prior studies.[29] While the strength of this study resides in the population-based approach, there are possible limitations that may decrease the generalizability of our findings. The clinical threshold at which an outpatient visit is considered to be indicated will likely vary between cultures and healthcare systems. This variation in clinical thresholds also exists for hospitalizations and ICU admissions, but is probably considerably less than for outpatient evaluation. In addition, the clinical severity score was loosely defined and was largely based on the clinical impression of the physician conducting the enrollment interview. Inter-rater variability was not quantified and we are therefore unable to speculate how this predictor will generalize to other settings. As the need for in-person evaluation at the COVID Outpatient Clinic was also determined by the same healthcare professionals during telephone interviews, it is perhaps not surprising that the clinical severity score was predictive. However, the clinical severity score during the enrollment interview also predicted the need for hospital admission, ICU care and death which downplays the possibility. Furthermore, a study examining the humoral response to SARS-CoV-2 in an overlapping cohort found a strong correlation between the clinical severity score and higher antibody levels, supporting the validity of the clinical severity score.[7]

    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.
    • No funding statement was detected.
    • 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 10.1101/2021.07.19.21260759v1 on medRxiv