[Skip to Navigation]
Sign In
Figure 1.  Odds Ratios of COVID-19 Convalescent Plasma (CCP) Efficacy and Expanded Treatment Benefit Index
Odds Ratios of COVID-19 Convalescent Plasma (CCP) Efficacy and Expanded Treatment Benefit Index

For all 6 outcomes, odds ratios of CCP efficacy (vs control) are shown as a function of the expanded treatment benefit index developed on the outcome of day-14 ordinal World Health Organization (WHO) scale. The plotted odds ratios (ORs) were estimated from cumulative proportional odds models or logistic models, depending on the outcome. The regressors were treatment, spline-represented treatment benefit index, and spline-represented treatment benefit index × treatment interaction, not adjusted for any other covariates. ORs for CCP efficacy of less than 1 indicate better outcome with CCP treatment than control. The cut points distinguishing benefit levels B1, B2, and B3 were 0.20 and 0.37 and are the same for all panels. The solid curves represent the ORs from the model, and the dashed curves indicate the associated 95% bootstrap confidence bands. The ORs for the 3 benefit levels are estimated from the primary bayesian models used in the analysis of the main results.

Figure 2.  Time to Death and Discharge Within 28 Days in 3 Benefit Level Groups
Time to Death and Discharge Within 28 Days in 3 Benefit Level Groups

Log-rank tests stratified for randomized clinical trials were used to compare COVID-19 Convalescent Plasma (CCP) and control for the mortality outcome. Gray competing risk test was used to compare CCP and control for the discharge outcome.

Figure 3.  Preexisting Health Status, Stage of COVID-19 Illness at Time of Treatment, and Benefit From COVID-19 Convalescent Plasma (CCP)
Preexisting Health Status, Stage of COVID-19 Illness at Time of Treatment, and Benefit From COVID-19 Convalescent Plasma (CCP)

Patients in the upper left corner (A), who have high preexisting risk but are at an early stage of COVID-19, are expected to have large benefit from CCP treatment. Patients with high preexisting risk who are at an advanced stage of COVID-19 (upper-right corner; B) as well as patients with low preexisting risk who are at early stage of COVID-19 (lower-left corner; C) are expected to benefit less from CCP. Patients with low preexisting risk who are at an advanced stage of COVID-19 (lower-right corner; D) are not expected to benefit and might experience harm from CCP treatment. WHO indicates World Health Organization.

Figure 4.  Predicted Patient Status for 4 Sample Patients
Predicted Patient Status for 4 Sample Patients

All 4 hypothetical patients were aged 60 years and had blood type O. Patient A had high preexisting risk (ie, cardiovascular disease, diabetes, and pulmonary disease) and early-stage COVID, with a treatment benefit index score of 0.85 (benefit level B1); patient B, high preexisting risk and later-stage COVID-19, with a treatment benefit index score of 0.68 (benefit level B1); patient C, low preexisting risk and early-stage COVID-19, with a treatment benefit index score of 0.36 (benefit level B2); and patient D, low preexisting risk and late-stage COVID-19, with a treatment benefit score of 0.19 (benefit level B3). CCP indicates COVID-19 convalescent plasma; WHO, World Health Organization.

Table.  Baseline Characteristics of Patients in the COMPILE Study by Benefit Level, Determined From the Expanded Treatment Benefit Index
Baseline Characteristics of Patients in the COMPILE Study by Benefit Level, Determined From the Expanded Treatment Benefit Index
Supplement.

eAppendix 1. TBI Models and Abbreviations

eAppendix 2. Workflow and Internal Cross-validation

eFigure 1. Workflow Diagram for Developing the TBI

eAppendix 3. Split-Sample-Simulation Cross-validation, 1000 replications

eAppendix 4. Leave-One-RCT-Out Cross-validation

eTable 1. RCTs Participating in COMPILE

eAppendix 5. Leave-One-Enrollment-Quarter-Out Cross-validation

eFigure 2. Median ORs and 95% Credible Intervals From the Posterior Probability Distributions for the ORs, by Patient Enrollment Quarters

eAppendix 6. Determination of 3 Levels of Benefit

eAppendix 7. Application of Other Methods for Treatment Classification Rules

eTable 2. Results in Terms of Odds Ratios from Cross-validation Based on 400 Splits Into Training Data Set and Validation Data Set

eTable 3. Results in Terms of Values From Cross-validation Based on 400 Splits Into Training Data Set and Validation Data Set

eAppendix 8. Baseline Characteristics of the Participants in the Study Used to Develop the TBIs

eTable 4. Baseline Characteristics of the Participants in the COMPILE Study by Treatments

eFigure 3. Missingness for Each of the Key Baseline Characteristics and Outcomes and Each Combination of the Variables for the 82 Participants Dropped From the Complete Case Analysis

eAppendix 9. Main Effect Model

eTable 5. Main Effects Model

eAppendix 10. Patient Features Considered for Inclusion in the TBI

eTable 6. Candidate Sets of Features Considered for Inclusion in the TBI

eAppendix 11. Results From 1000 Split-Sample-Simulation Cross-validation

eTable 7. Results From 1000 Split-Sample Simulation Cross-validation

eFigure 4. Subgroup-Specific OR Distributions Obtained From a 1000 Split-Sample Simulation Cross-validation

eTable 8. Value Results From the 1000 Split-Sample Simulation Cross-validation

eAppendix 12. Results From Leave-One-RCT-Out Cross-validation

eTable 9. Results From a Leave-One-RCT-Out Cross-validation

eTable 10. Value Results From a Leave-One-RCT-Out Cross-validation

eAppendix 13. Results From Leave-One-Enrollment-Quarter-Out Cross-validation

eTable 11. Results From a Leave-One-Enrollment-Quarter-Out Cross-validation

eTable 12. Results From a Leave-One-Enrollment-Quarter-Out Cross-validation, Before Averaging Across the 4 Enrollment Quarters (Q2-Q5) for the Basic and Expanded TBIs

eTable 13. Value Results From a Leave-One-Enrollment-Quarter-Out Cross-validation

eAppendix 14. Comparisons With a Naive Treatment Decision Approach

eFigure 5. Distributions of Subgroup-Specific ORs Obtained From 1000 Randomly Split Testing Sets for the Ordinal Outcome WHO Score at Day 14

eFigure 6. Distributions of Subgroup-Specific ORs Obtained From 1000 Randomly Split Testing Sets for the Ordinal Outcome WHO Score at Day 28

eFigure 7. Distributions of Subgroup-Specific ORs Obtained From 1000 Randomly Split Testing Sets for Binary Outcome of WHO Score of at Least 7 at day 14

eFigure 8. Distributions of Subgroup-Specific ORs Obtained From 1000 Randomly Split Testing Sets for Binary Outcome of WHO Score of at Least 7 at Day 28

eFigure 9. Distributions of Subgroup-Specific ORs Obtained From 1000 Randomly Split Testing Sets for Death at Day 14

eFigure 10. Distributions of Subgroup-Specific ORs Obtained From 1000 Randomly Split Testing Sets for Death at Day 28

eAppendix 15. Specification of the Basic and Expanded TBIs

eTable 14. Coefficients of the Linear Combination for Baseline Patient Characteristics in the Basic and Expanded TBIs

eAppendix 16. Cross-Stratification Table Comparing the Basic and Expanded TBIs

eTable 15. Benefit Stratification Table

eAppendix 17. Proportional Odds Assumption Test

eTable 16. Estimates of Coefficients Associated With HTE

eTable 17. Brant Test Results

eAppendix 18. Three Levels of Benefit for the Basic and Expanded TBIs

eFigure 11. Basic TBI Model-Based Cumulative Odds Ratio for the Ordinal WHO Score at Day 14, as a Function of the Basic TBI

eFigure 12. Expanded TBI Model-Based Cumulative Odds Ratio for the Ordinal WHO Score at Day 14, as a Function of the Expanded TBI

eTable 18. Odds Ratios Associated with All 6 Outcomes, Under 3 Levels of Categorization From the Basic and Expanded TBIs

eTable 19. Results Regarding the Expected Incidence (Value) of the 4 Clinically Undesirable Binary Outcomes When CCP is Administered to the 3 Patient Subgroups Identified by the Basic and Expanded TBIs

eAppendix 19. Baseline Patient Characteristics by Benefit Levels

eTable 20. Baseline Characteristics of COMPILE Patients in the 3 Benefit-Level Groups Defined From the Expanded TBI

eFigure 13. Proportions of the 3 Benefit-Level Groups Identified From the Expanded TBI, by Quarter

eTable 21. Baseline Patient Characteristics for the Subgroup of Patients Who Would and Would Not Be Advised to Receive CCP, Determined by the Expanded TBI

eAppendix 20. CCP Efficacy in the 3 Benefit Groups With Respect to All 6 Outcomes, Determined by the Expanded TBI

eTable 22. Posterior Probability Distributions of the ORs of CCP vs Control for All Outcomes, by Benefit Group Determined From the Expanded TBI

eFigure 14. Forest Plots of Median ORs and 95% Credible Intervals From the Posterior Probability Distributions for the ORs by Benefit Groups Determined From the Expanded TBI for All 6 Outcomes

eAppendix 21. Differential Treatment Effect by Benefit Levels, in Comparison With Differential Treatment Effect by Single Patient Characteristics

eFigure 15. Odds Ratios and 95% Credible Intervals for Subgroup-Specific Odds for Individual Baseline Covariates and 3 Benefit Groups for WHO Score on Day 14

eFigure 16. Odds Ratios and 95% Credible Intervals for Subgroup-Specific Odds for Individual Baseline Covariates and 3 Benefit Groups for WHO Score 7 to 10 on Day 14

eFigure 17. Odds Ratios and 95% Credible Intervals for Subgroup-Specific Odds for Individual Baseline Covariates and 3 Benefit Groups for WHO Score on Day 28

eFigure 18. Odds Ratios and 95% Credible Intervals for Subgroup-Specific Odds for Individual Baseline Covariates and 3 Benefit Groups for WHO Score 7 to 10 on Day 28

eFigure 19. Odds Ratios and 95% Credible Intervals for Subgroup-Specific Odds for Individual Baseline Covariates and 3 Benefit Groups for Mortality on Day 14

eFigure 20. Odds Ratios and 95% Credible Intervals for Subgroup-Specific Odds for Individual Baseline Covariates and 3 Benefit Groups for Mortality on Day 28

eAppendix 22. CCP Benefit as a Function of the Basic TBI

eFigure 21. CCP Benefit as a Function of the Basic TBI

eAppendix 23. Baseline Patient Characteristics by Benefit Levels Determined From the Basic TBI

eTable 23. Baseline Characteristics of COMPILE Patients in the 3 Benefit-Level Groups Defined by the Basic TBI

eFigure 22. Proportions of the 3 Benefit Groups (B1, B2 and B3) Identified From the Basic TBI, by Quarter

eTable 24. Baseline Patient Characteristics for the Subgroup of Patients Who Would and Would Not Be Advised to Receive CCP, Determined by the Basic TBI

eAppendix 24. CCP Efficacy in the 3 Benefit Groups With Respect to All 6 Outcomes, Determined by the Basic TBI

eTable 25. Posterior Probability Distributions of the Odds Ratios of CCP vs Control for All Outcomes, by Benefit Group Determined From the Basic TBI

eFigure 23. Forest Plots of Median Odds Ratios and 95% Credible Intervals From the Posterior Probability Distributions for the Odds Ratios by Benefit Groups Determined From the Basic TBI for All 6 Outcomes

eAppendix 25. Kaplan-Meier Plots for Time to Death and Time to Hospital Discharge, With Log-Rank Test for Time to Death and Gray Competing Risk Test for Time to Hospital Discharge, Basic TBI

eFigure 24. Comparison Between CCP and Control Patients for Time to Death and Time to Hospital Discharge by Benefit Level, Determined by the Basic TBI

eAppendix 26. Validation on Expanded Access Program (EAP) Data

eTable 26. Baseline Characteristics for Participants Treated Under the EAP With Available Data for All Specified Parameters

eTable 27. Summary of Balance for Matched Data for EAP

eTable 28. Summary of Balance for Matched Data for EAP, High Titer Only

eTable 29. Participants in the Matched Data, by Benefit Levels Defined by the Expanded and the Basic TBIs and Depending on Whether All EAP Patients or Only Those Treated With High-Titer Plasma Were Used For Validation

eTable 30. CCP Benefit Odds Ratios Within Different Benefit Levels Defined by the Basic and the Expanded TBIs and Depending on Whether Subgrouping Was Restricted to Patients with High-Titer CCP

eAppendix 27. Validation on an Emergency Use Authorization (EUA) Study

eTable 31. Baseline Characteristics of CCP Recipients at Montefiore Medical Center, Bronx, NY

eTable 32. Summary of Balance for Matched Data for the EUA

eTable 33. Patients in Different Benefit Levels defined by the Basic TBI and the Expanded TBI for the EUA

eTable 34. Subgroup-Specific CCP Benefit Odds Ratios for Benefit Levels Defined by the Basic TBI and the Expanded TBI for the EUA

eAppendix 28. Validation on External RCT 1

eTable 35. Baseline Patient Characteristics and Clinical Information of the First RCT

eTable 36. Major Pharmacotherapy as the Standard of Care

eTable 37. Patients in Different Benefit Levels Based on the Basic TBI in the First RCT

eTable 38. Subgroup-Specific CCP Benefit Odds Ratios With Respect to 30-Day Mortality, by Benefit Level Defined by the Basic TBI, in the First RCT

eFigure 25. CCP vs Control as a Function of the Basic TBI (Which Does Not Require Blood Type Information), With the 3 Benefit Groups Overlaid, Evaluated on the First External RCT

eAppendix 29. Validation on External RCT 2

eTable 39. Baseline Patient Characteristics for the Second RCT

eTable 40. Patients in Different Benefit Levels Based on the Expanded and Basic TBIs for the Second RCT

eTable 41. CCP Benefit Odds Ratios for Ordinal Outcomes at Days 14 and 30 for the Benefit Groups Defined by the Basic and Expanded TBIs, Evaluated on the Second External RCT

eFigure 26. CCP Benefit Odds Ratios for WHO 6-Point Scale Ordinal Outcomes at Days 14 and 30, as Functions of the Basic and Expanded TBIs, Evaluated on the Second External RCT

eTable 42. CCP Benefit Odds Ratios for the Binary Outcome of Ventilation or Worse at Days 14 and 30 for Benefit Groups Defined by the Basic and Expanded TBIs, Evaluated on the Second External RCT

eFigure 27. CCP Benefit Odds Ratios for Binary Outcome of Ventilation or Worse at Days 14 and 30 for Benefit Groups Defined by the Basic and Expanded TBIs, Evaluated on the Second External RCT

eAppendix 30. More Results From the Expanded TBI

eFigure 28. Probability of the 11-Point WHO Ordinal Scales Categorized into 4 Ordinal Categories, as a Function of the Expanded TBI

eAppendix 31. Response Surface and Scatterplots Associated With the Expanded TBI Model

eFigure 29. Fitted POM

eFigure 30. Fitted POM Displayed on the Probability Scale

eFigure 31. Scatterplot of the Risk Index and the TBI in the Fitted POM

eFigure 32. Scatterplot of the Risk Index and the TBI-by-Treatment Interaction Term in the Fitted POM

eFigure 33. Empirical Cumulative Distribution Functions for the Basic and Expanded TBIs With the Associated Benefit Levels Overlaid

1.
Tonelli  MR, Shirts  BH.  Knowledge for precision medicine: mechanistic reasoning and methodological pluralism.   JAMA. 2017;318(17):1649-1650. doi:10.1001/jama.2017.11914PubMedGoogle ScholarCrossref
2.
Blackstone  EH.  Precision medicine versus evidence-based medicine: individual treatment effect versus average treatment effect.   Circulation. 2019;140(15):1236-1238. doi:10.1161/CIRCULATIONAHA.119.043014PubMedGoogle ScholarCrossref
3.
DeMerle  K, Angus  DC, Seymour  CW.  Precision medicine for COVID-19: phenotype anarchy or promise realized?   JAMA. 2021;325(20):2041-2042. doi:10.1001/jama.2021.5248PubMedGoogle ScholarCrossref
4.
Bzdok  D, Varoquaux  G, Steyerberg  EW.  Prediction, not association, paves the road to precision medicine.   JAMA Psychiatry. 2021;78(2):127-128. doi:10.1001/jamapsychiatry.2020.2549PubMedGoogle ScholarCrossref
5.
Zhao  Y, Zeng  D, Rush  AJ, Kosorok  MR.  Estimating individualized treatment rules using outcome weighted learning.   J Am Stat Assoc. 2012;107(449):1106-1118. doi:10.1080/01621459.2012.695674PubMedGoogle Scholar
6.
Song  R, Kosorok  M, Zeng  D, Zhao  Y, Laber  E, Yuan  M.  On sparse representation for optimal individualized treatment selection with penalized outcome weighted learning.   Stat. 2015;4(1):59-68. doi:10.1002/sta4.78PubMedGoogle ScholarCrossref
7.
Laber  EB, Zhao  YQ.  Tree-based methods for individualized treatment regimes.   Biometrika. 2015;102(3):501-514. doi:10.1093/biomet/asv028PubMedGoogle ScholarCrossref
8.
Petkova  E, Tarpey  T, Su  Z, Ogden  RT.  Generated effect modifiers (GEM’s) in randomized clinical trials.   Biostatistics. 2017;18(1):105-118. doi:10.1093/biostatistics/kxw035PubMedGoogle ScholarCrossref
9.
Liu  Y, Wang  Y, Kosorok  MR, Zhao  Y, Zeng  D.  Augmented outcome-weighted learning for estimating optimal dynamic treatment regimens.   Stat Med. 2018;37(26):3776-3788. doi:10.1002/sim.7844PubMedGoogle ScholarCrossref
10.
Park  H, Petkova  E, Tarpey  T, Ogden  RT.  A single-index model with multiple-links.   J Stat Plan Inference. 2020;205:115-128. doi:10.1016/j.jspi.2019.05.008PubMedGoogle ScholarCrossref
11.
Park  H, Petkova  E, Tarpey  T, Ogden  RT.  A constrained single-index regression for estimating interactions between a treatment and covariates.   Biometrics. 2021;77(2):506-518.PubMedGoogle ScholarCrossref
12.
Petkova  E, Antman  EM, Troxel  AB.  Pooling data from individual clinical trials in the COVID-19 era.   JAMA. 2020;324(6):543-545. doi:10.1001/jama.2020.13042PubMedGoogle ScholarCrossref
13.
Troxel  AB, Petkova  E, Goldfeld  K,  et al.  Association of convalescent plasma treatment with clinical status in patients hospitalized with COVID-19: a meta-analysis.   JAMA Netw Open. 2022;5(1):e2147331. doi:10.1001/jamanetworkopen.2021.47331Google Scholar
14.
Goldfeld  K, Wu  D, Tarpey  T,  et al.  Prospective individual patient data meta-analysis: Evaluating convalescent plasma for COVID-19.   Stat Med. 2021;40(24):5131-5151. doi:10.1002/sim.9115PubMedGoogle ScholarCrossref
15.
WHO Working Group on the Clinical Characterisation and Management of COVID-19 infection.  A minimal common outcome measure set for COVID-19 clinical research.   Lancet Infect Dis. 2020;20(8):e192-e197. doi:10.1016/S1473-3099(20)30483-7PubMedGoogle ScholarCrossref
16.
Park  H, Petkova  E, Tarpey  T, Ogden  RT.  A single-index model with a surface-link for optimizing individualized dose rules.   J Computational Graphical Stat. Published online June 21, 2021. doi:10.1080/10618600.2021.1923521Google Scholar
17.
Gray  RJ.  A class of k-sample tests for comparing the cumulative incidence of a competing risk.   Ann Stat. 1988;16(3):1141-1154. Accessed December 20, 2021. https://www.jstor.org/stable/2241622Google ScholarCrossref
18.
McCaw  ZR, Tian  L, Vassy  JL,  et al.  How to quantify and interpret treatment effects in comparative clinical studies of COVID-19.   Ann Intern Med. 2020;173(8):632-637. doi:10.7326/M20-4044PubMedGoogle ScholarCrossref
19.
Janes  H, Pepe  MS, Gu  W.  Assessing the value of risk predictions by using risk stratification tables.   Ann Intern Med. 2008;149(10):751-760. doi:10.7326/0003-4819-149-10-200811180-00009PubMedGoogle ScholarCrossref
20.
Senefeld  JW, Johnson  PW, Kunze  KL,  et al.  Program and patient characteristics for the United States Expanded Access Program to COVID-19 convalescent plasma.   medRxiv. Preprint posted online April 11, 2021. doi:10.1101/2021.04.08.21255115Google Scholar
21.
Joyner  MJ, Carter  RE, Senefeld  JW,  et al.  Convalescent plasma antibody levels and the risk of death from COVID-19.   N Engl J Med. 2021;384(11):1015-1027. doi:10.1056/NEJMoa2031893PubMedGoogle ScholarCrossref
22.
US Food and Drug Administration. FDA issues Emergency Use Authorization for convalescent plasma as potential promising COVID-19 treatment, another achievement in administration’s fight against pandemic. April 23, 2020. Accessed December 20, 2021. https://www.fda.gov/news-events/press-announcements/fda-issues-emergency-use-authorization-convalescent-plasma-potential-promising-covid-19-treatment
23.
Ray  Y, Paul  SR, Bandopadhyay  P,  et al.  Clinical and immunological benefits of convalescent plasma therapy in severe COVID-19: insights from a single center open label randomised control trial.   medRxiv. Preprint posted online November 29, 2020. doi:10.1101/2020.11.25.20237883Google Scholar
24.
Simonovich  VA, Burgos Pratx  LD, Scibona  P,  et al; PlasmAr Study Group.  A randomized trial of convalescent plasma in COVID-19 severe pneumonia.   N Engl J Med. 2021;384(7):619-629. doi:10.1056/NEJMoa2031304PubMedGoogle ScholarCrossref
25.
Madhi  SA, Baillie  V, Cutland  CL,  et al; NGS-SA Group; Wits-VIDA COVID Group.  Efficacy of the ChAdOx1 nCoV-19 COVID-19 Vaccine against the B.1.351 variant.   N Engl J Med. 2021;384(20):1885-1898. doi:10.1056/NEJMoa2102214PubMedGoogle ScholarCrossref
26.
Park H, Tarpey T, Li Y, et al. Convalescent plasma treatment benefit index calculator. January 21, 2022. http://covid-convalescentplasma-tbi-calc.org
2 Comments for this article
EXPAND ALL
Link to the CCP TBI calculator
Eva Petkova, PhD | New York University
For a couple of hours the link to the COVID-19 convalescent plasma (CCP) treatment benefit index (TBI) calculator developed in this paper was inactive. Now this has been corrected and the link works
http://covid-convalescentplasma-tbi-calc.org
CONFLICT OF INTEREST: I am the corresponding author for this paper
Blood group and clinical outcomes
Neil Blumberg, MD | University of Rochester
One possible reason for the better outcomes with recipients of blood group A and AB is the common practice of transfusing these types of plasma to group O recipients as "compatible" and, in the case of AB "universal donor." Both randomized trials and observational studies suggest that use of ABO mismatched or "universal donor" plasma increases morbidity and mortality, contrary to a century of dogma. There are no good data to support use of universal donor AB plasma to non-AB patients.

Group O patients have large amounts of anti-A and anti-B and would be expected
to be harmed by A and AB plasma whereas group A and AB patients would not have the formation of immune complexes of soluble A and B antigen that would occur more frequently in group O recipients. Thus it is possible that the signal that A and AB patients are benefited may be a result of the relative harm to group O and, to a lesser extent, group B patients.
CONFLICT OF INTEREST: None Reported
READ MORE
Original Investigation
Infectious Diseases
January 25, 2022

Development and Validation of a Treatment Benefit Index to Identify Hospitalized Patients With COVID-19 Who May Benefit From Convalescent Plasma

Author Affiliations
  • 1Division of Biostatistics, Department of Population Health, New York University Grossman School of Medicine, New York
  • 2Department of Environmental Medicine, New York University Grossman School of Medicine, New York
  • 3Department of Biostatistics, School of Public Health, University of Washington, Seattle
  • 4Biostatistics and Research Decision Sciences, Merck Research Labortory, Merck & Co Inc, Rahway, New Jersey
  • 5Translational Research Unit of Excellence, Council Of Scientific And Industrial Research–Indian Institute of Chemical Biology, Kolkata, India
  • 6Infectious Disease, Beleghata General Hospital, Kolkata, India
  • 7School of Tropical Medicine, Kolkata, India
  • 8Medical College Hospital, Kolkata, India
  • 9Center for Biologics Evaluation and Research, Office of Biostatistics and Epidemiology, Analytics and Benefit-Risk Assessment Team, US Food and Drug Administration, Silver Spring, Maryland
  • 10Office of Blood Research and Review, Center for Biologics Evaluation and Research, US Food and Drug Administration, Silver Spring, Maryland
  • 11Albert Einstein College of Medicine and Montefiore Medical Center, Bronx, New York
  • 12Indian Council of Medical Research, New Delhi, India
  • 13Clinical Pharmacology Section, Department of Internal Medicine and Department of Research, Hospital Italiano de Buenos Aires, Buenos Aires, Argentina
  • 14Clinical Pharmacology Section, Internal Medicine Service, Hospital Italiano de Buenos Aires, Buenos Aires, Argentina
  • 15Transfusional Medicine Service, Hospital Italiano de Buenos Aires, Buenos Aires, Argentina
  • 16Department of Research, Hospital Italiano de Buenos Aires, Buenos Aires, Argentina
  • 17Hospital Universitario Puerta de Hierro Majadahonda, Madrid, Spain
  • 18Department of Medicine, University of Pennsylvania Perelman School of Medicine, Philadelphia
  • 19Zuckerberg San Francisco General, University of California, San Francisco
  • 20Department of Intensive Care Medicine, University Hospitals Leuven, Leuven, Belgium
  • 21Hospital Universitário de Brasília, University of Brasília, Brasília, Brazil
  • 22Departments of Medicine and Microbiology, New York University Grossman School of Medicine, New York
  • 23Department of Internal Medicine, Section of Infectious Diseases, Erasmus University Medical Center, Rotterdam, the Netherlands
  • 24Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts
  • 25Department of Child and Adolescent Psychiatry, New York University Grossman School of Medicine
  • 26Nathan S. Kline Institute for Psychiatric Research, Orangeburg, New York
JAMA Netw Open. 2022;5(1):e2147375. doi:10.1001/jamanetworkopen.2021.47375
Key Points

Question  What patient characteristics are associated with benefit from treatment with COVID-19 convalescent plasma (CCP)?

Findings  This prognostic study of 2287 patients hospitalized with COVID-19 identified a combination of baseline characteristics that predict a gradation of benefit from CCP compared with treatment without CCP. Preexisting health conditions (diabetes, cardiovascular and pulmonary diseases), blood type A or AB, and earlier stage of COVID-19 were associated with a larger treatment benefit.

Meaning  These findings suggest that simple patient information collected at hospitalization can be used to guide CCP treatment decisions for patients with COVID-19.

Abstract

Importance  Identifying which patients with COVID-19 are likely to benefit from COVID-19 convalescent plasma (CCP) treatment may have a large public health impact.

Objective  To develop an index for predicting the expected relative treatment benefit from CCP compared with treatment without CCP for patients hospitalized for COVID-19 using patients’ baseline characteristics.

Design, Setting, and Participants  This prognostic study used data from the COMPILE study, ie, a meta-analysis of pooled individual patient data from 8 randomized clinical trials (RCTs) evaluating CCP vs control in adults hospitalized for COVID-19 who were not receiving mechanical ventilation at randomization. A combination of baseline characteristics, termed the treatment benefit index (TBI), was developed based on 2287 patients in COMPILE using a proportional odds model, with baseline characteristics selected via cross-validation. The TBI was externally validated on 4 external data sets: the Expanded Access Program (1896 participants), a study conducted under Emergency Use Authorization (210 participants), and 2 RCTs (with 80 and 309 participants).

Exposure  Receipt of CCP.

Main Outcomes and Measures  World Health Organization (WHO) 11-point ordinal COVID-19 clinical status scale and 2 derivatives of it (ie, WHO score of 7-10, indicating mechanical ventilation to death, and WHO score of 10, indicating death) at day 14 and day 28 after randomization. Day 14 WHO 11-point ordinal scale was used as the primary outcome to develop the TBI.

Results  A total of 2287 patients were included in the derivation cohort, with a mean (SD) age of 60.3 (15.2) years and 815 (35.6%) women. The TBI provided a continuous gradation of benefit, and, for clinical utility, it was operationalized into groups of expected large clinical benefit (B1; 629 participants in the derivation cohort [27.5%]), moderate benefit (B2; 953 [41.7%]), and potential harm or no benefit (B3; 705 [30.8%]). Patients with preexisting conditions (diabetes, cardiovascular and pulmonary diseases), with blood type A or AB, and at an early COVID-19 stage (low baseline WHO scores) were expected to benefit most, while those without preexisting conditions and at more advanced stages of COVID-19 could potentially be harmed. In the derivation cohort, odds ratios for worse outcome, where smaller odds ratios indicate larger benefit from CCP, were 0.69 (95% credible interval [CrI], 0.48-1.06) for B1, 0.82 (95% CrI, 0.61-1.11) for B2, and 1.58 (95% CrI, 1.14-2.17) for B3. Testing on 4 external datasets supported the validation of the derived TBIs.

Conclusions and Relevance  The findings of this study suggest that the CCP TBI is a simple tool that can quantify the relative benefit from CCP treatment for an individual patient hospitalized with COVID-19 that can be used to guide treatment recommendations. The TBI precision medicine approach could be especially helpful in a pandemic.

Introduction

Participants in randomized clinical trials (RCTs) typically exhibit heterogeneity of the treatment effect (HTE) of tested interventions. The traditional approach of focusing on the average effect has important limitations when making clinical decisions for individual patients.1-4 Precision medicine approaches have been developed to identify individual patients most likely to benefit from specific therapies.5-9

In this article, we report on an investigation to discover profiles of patients with COVID-19 associated with different benefit from COVID-19 convalescent plasma (CCP) treatment. The approach is based on a treatment benefit index (TBI), a continuous measure defined as a combination of patient characteristics that maximizes its interaction with CCP treatment.10,11 The TBI was derived using the COMPILE study12 and was validated in multiple external data sets.

The COMPILE study pooled individual patient data from 8 international RCTs and found no overall association between CCP and patient outcomes.13 The precision medicine investigations were prespecified in the COMPILE study’s statistical analysis plan.14

HTE

Heterogeneity of CCP treatment benefit in COMPILE was observed with respect to (1) outcomes (clinical status based on the WHO ordinal scale15; binary outcomes of mechanical ventilation or death and all-cause mortality); (2) timing of assessments (day 14 vs 28); (3) quarter of enrollment (April to June 2020, July to September 2020, October to December 2020, and January to March 2021); and (4) patient demographic and clinical characteristics (eg, age, sex, comorbid medical conditions). While the first 3 factors correspond to nonpatient-related sources, factor 4 reflects patient-related heterogeneity that the TBI was constructed to identify. Quarter of enrollment, possibly through the evolving standard of care, might also affect the relative efficacy of CCP for patients with the same profile, but in different time periods; therefore, we also investigated the potential influence of quarter of enrollment on changes in the TBIs.

TBI Objectives

The goal of this study was to guide CCP treatment recommendations by providing an estimate of a differential treatment outcome when a patient is treated with CCP vs without CCP. A larger differential in favor of CCP would indicate a more compelling reason for recommending CCP.

Two objectives were balanced: simplicity in terms of patient characteristics for implementation and accuracy in terms of benefit prediction for individual patients. Recognizing that not all patient information might be available when treatment decisions need to be made urgently, a complementary goal became the development of a basic TBI, on which improvements are possible with additional information. To demonstrate this idea, a basic TBI, using only easily obtained characteristics not including blood type, was derived, and then an expanded TBI, augmented with blood type information, was developed, improving the benefit prediction.

Methods
Data for TBI Development

COMPILE included 2369 hospitalized adults, not receiving mechanical ventilation at randomization, enrolled April 2020 to March 2021 (Table). COMPILE was approved by the New York University institutional review board, which determined that the study did not involve human participants because it used only deidentified data, thereby waiving the requirement for informed consent. This study followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline. The control treatments varied across RCTs: standard of care (SOC), SOC plus saline, and SOC plus nonconvalescent plasma. While a few patients were not treated according to the randomization (due to administrative errors), the TBI derivation used treatment as randomized.

Six end points were available. There were 3 outcomes (ie, the ordinal WHO 11-point scale,15 a binary indicator of WHO score 7 to 10 [receiving mechanical ventilation or death], and binary indicator of WHO score 10 [mortality]) at 2 assessment times (ie, 14 ± 1 and 28 ± 2 days post randomization [hereafter day 14 and day 28]).13 The ordinal WHO scores and the indicators for mechanical ventilation or death at day 14 were coprimary outcomes in COMPILE. The TBIs were developed on the ordinal day-14 WHO score and then tested on all other outcomes.

Statistical Analysis
Deriving the TBIs

The TBIs were developed using a single-index regression for estimating interactions between treatment and covariates,11 extended to accommodate ordinal outcomes.16 The TBIs are linear combinations of baseline characteristics constructed to optimally differentiate the association of CCP treatment with outcomes on day-14 WHO scores from that of control, using cumulative proportional odds models16 (POM; eAppendix 1 and eAppendix 2 in the Supplement). Candidate TBIs with different sets of baseline characteristics were identified from extensive internal cross-validation to optimize the generalizability of the TBIs (eFigure 1 in the Supplement), and the selected basic and expanded TBIs were tested using external data. Several forms of cross-validation were used: cross-validation from multiple random splits of the whole sample into training and testing sets (split-sample simulation; eAppendix 3 in the Supplement); cross-validation based on different RCTs (leave-one-RCT-out) to assess generalizability across RCTs (eAppendix 4 and eTable 1 in the Supplement); and cross-validation based on enrollment quarters (leave-one-enrollment-quarter-out) to assess stability of the performance over time (eAppendix 5 and eFigure 2 in the Supplement). Final sets of baseline characteristics were specified for the basic and expanded TBIs, and the associated POMs were reestimated from the whole data to give the final coefficients for the TBIs.

Utility Evaluation

In the derivation cohort, performance was measured by the CCP benefit in 2 subgroups identified from POM, one expected to benefit (B) and one not expected to benefit (NB) from CCP, in terms of their subgroup-specific odds ratios (ORs) and their ratios (ie, OR for B divided by OR for NB) to measure the difference between the CCP benefit in B vs NB. The cut point is where the CCP and the control curves crossed. An OR of less than 1 indicates CCP efficacy, with a smaller ratio of ORs indicating better TBI performance. An additional performance measure was the value, defined for the binary outcomes as the expected proportion of patients with the outcome if individuals in B are treated with CCP and those in NB are treated with the control; lower values were preferable since the outcomes are undesirable (eAppendix 2 in the Supplement).

The TBIs range from 0 to 1: larger values are associated with larger CCP benefit. For clinical utility, we operationalized the continuous TBIs as 3 benefit levels: large benefit (B1), modest benefit (B2), and no benefit or potential harm (B3) (eAppendix 6 in the Supplement). We evaluated the TBIs based on the within–benefit level CCP efficacy ORs. While the TBIs were derived on the ordinal WHO score at day 14, efficacy was assessed with respect to all 6 outcomes. ORs were obtained from models used in the main COMPILE analysis13,14: Bayesian POMs and logistic regressions, adjusted for the same covariates as in the COMPILE analysis with the same prior distributions. Bayesian posterior distributions of the respective ORs were obtained for each benefit level. Tests were 2-tailed. We contrasted the utility of the TBIs vs that of the individual baseline variables prespecified in COMPILE with outcomes to assess the advantage of the TBIs for guiding clinical decisions as an alternative to using individual baseline covariates. Finally, we contrasted the benefit levels with respect to the comparison of CCP vs control on time to all-cause mortality (log-rank test, stratified by RCT) and time to discharge within 28 days (Gray competing risk analysis17) to test for differences between the cumulative incidence functions.18

Alternative Development Methods

Fourteen alternative methods for developing treatment decision rules were used in search of a superior characterization of the HTE (eAppendix 7, eTable 2, and eTable 3 in the Supplement). However, none of them outperformed the TBI approach.

External Validation

Four CCP data sets (single-arm and RCTs) external to COMPILE were used to test the selected basic and expanded TBIs. The validation was based on the ordering of the ORs in the 3 benefit levels (ie, the B1 group having the smallest ORs and the B3 group having the largest would constitute validation). All data processing, analysis, and visualization were performed in R version 2021 (R Project for Statistical Computing). For frequentist inference, statistical significance was set at α = .05, and all tests were 2-tailed.

Results
TBIs Derivation

Baseline characteristics of the 2369 participants of the COMPILE study appear in eAppendix 8 and eTable 4 in the Supplement. The median (IQR) age was 60 (50-72) years, and 845 participants (35.7%) were women. For the day-14 ordinal WHO score, the posterior median CCP efficacy OR was 0.94 (95% credible interval [CrI], 0.74-1.19). Details appear in Troxel et al.13 The TBIs were developed on 2287 complete cases. The marginal and joint distributions of missing data appear in eFigure 3 in the Supplement. In the derivation cohort, the mean (SD) age was 60.3 (15.2) years, and 815 participants (35.6%) were women. The covariates’ main effect coefficients’ portion of model development, which is not a part of the TBI, appears in eAppendix 9 and eTable 5 in the Supplement. To identify variables included in the interaction parts of POM that define the TBIs, we considered 24 combinations of baseline characteristics (eAppendix 10 and eTable 6 in the Supplement). Results from split-sample cross-validation (eAppendix 11, eTable 7, eTable 8, and eFigure 4 in the Supplement), leave-one-RCT-out cross-validation (eAppendix 12, eTable 9, and eTable 10 in the Supplement), and leave-one-enrollment-quarter-out cross-validation (eAppendix 13 and eTables 11-13 in the Supplement) identified the basic and expanded TBIs. Additional results from the split-sample cross-validation investigations appear in eAppendix 14 and eFigures 5 to 10 in the Supplement.

eAppendix 15 and eTable 14 in the Supplement report the coefficients and 95% bootstrap CIs of the linear combinations that define the basic and expanded TBIs. The improvement in the benefit prediction by TBI because of the inclusion of information on blood type can be assessed by comparing the basic and expanded TBIs with respect to the OR ratios (eTables 7, 9, 11, and 12 in the Supplement) and value (eTables 8, 10, and 13 in the Supplement). Additional comparison based on cross-classification19 appears in eAppendix 16 and eTable 15 in the Supplement. The proportional odds assumption for POM was assessed to be reasonable (eAppendix 17 and eTables 16 and 17 in the Supplement). Additional information on the fitted POM is given in eAppendices 30 and 31 and eFigures 28 to 33 in the Supplement.

Major Findings From the Internal Cross-Validation

The TBIs were assessed through leave-one-RCT-out cross-validation (eTable 1 in the Supplement for the 8 RCTs), which provides support for their generalizability (eTables 9 and 10 in the Supplement). The TBIs’ performance in the leave-one-enrollment-quarter-out cross-validation (eTables 11-13 in the Supplement) indicated that although the efficacy of SOC changed over time, the relative benefit from CCP was determined by the same combination of patient characteristics.

Internal Evaluation of CCP TBI

Figure 1 shows the expanded TBI (developed on day-14 ordinal WHO scores) plotted against the unadjusted ORs for CCP efficacy for all 6 outcomes. All panels show a monotonically decreasing trend of the ORs (indicating an increase in the CCP benefit) as the TBI score increases from 0 to 1. Some of the OR curves and the 95% confidence bands exceeded 1 for very small TBI values, suggesting the possibility of harm from CCP as TBI approaches 0.

The dotted vertical lines in Figure 1 mark the groupings corresponding to large benefit (B1), modest benefit (B2), and no benefit/potential harm (B3) groups that operationalize the TBI for clinical utility. The benefit levels were determined by the day-14 ordinal WHO score, and the cut points were chosen to optimize the 3-category model fit as described in eAppendix 6 in the Supplement. eAppendix 18, eFigures 11 and 12, and eTables 18 and 19 in the Supplement contain more details on this categorization and the internal cross-validation results. The in-sample proportion of patients in these 3 groups were as follows: 629 participants (27.5%) in B1; 953 participants (41.7%), B2; and 705 (30.8%), B3. These same cut points and benefit levels are used for all panels of Figure 1 and in all analyses that follow.

The Table describes the distribution of patient characteristics by the 3 benefit levels (eAppendix 19, eTables 20 and 21, and eFigure 13 in the Supplement). The benefit level–specific CCP efficacy ORs; the posterior probability of an OR of less than 1, indicating the evidence of any CCP benefit; and the posterior probability of an OR of less than 0.80, indicating probability of more than minimal CCP benefit for all 6 outcomes appear in eAppendix 20, eTable 22, and eFigure 14 in the Supplement. With respect to day-14 ordinal WHO scores, the B1 group had a posterior median OR of 0.69 (95% CrI, 0.48-1.06), a posterior probability of any CCP benefit of 96%, and the posterior probability of more than minimal benefit of 77%, indicating strong evidence for benefit from CCP; for B2, the posterior median OR was 0.82 (95% CrI, 0.61-1.11), and posterior probabilities of any benefit of 90% and more than minimal benefit of 42%, indicating modest evidence for CCP efficacy; and for B3, the posterior median OR of 1.58 (95% CrI, 1.14-2.17) and zero posterior probabilities for any CCP benefit and more than minimal benefit, indicating strong evidence for no efficacy and potential harm. For all outcomes, patients in B1 were expected to have large benefit, patients in B2 expected to have modest benefit, and patients in B3 expected to have at least a potential for harm from CCP. Additionally, eAppendix 21 and eFigures 15 to 20 in the Supplement include forest plots of ORs for subgroups defined by individual patient characteristics for the 6 outcomes. The benefit levels defined by the TBI show stronger separation of ORs across all outcomes than any individual covariate.

Figure 2 summarizes results for time to death and time to hospital discharge (up to day 28). There was a large benefit with respect to mortality in B1, modest benefit in B2, and no benefit in B3. With respect to time to discharge, patients in B1 were discharged 2.3 (95% CI, 0.7-3.8) days earlier if treated with CCP compared with control treatment. Results for the basic TBI were similar (eAppendices 22-25, eFigures 21-24, and eTables 23-25 in the Supplement).

Figure 3 provides a visual interpretation of the TBI. The vertical axis corresponds to preexisting health risk and the horizontal axis corresponds to the stage of COVID-19 at time of treatment. CCP benefit depends on both: CCP is most associated with benefit for patients with high preexisting risk who have early-stage COVID-19 at time of treatment (ie, patients in the upper-left corner of the figure), and it is least associated with benefit—and potentially associated with harm—for patients with low preexisting risk and an advanced stage of COVID-19. Figure 4 shows 4 hypothetical patients with different preexisting health risks and stages of the disease at time of treatment; these patients roughly correspond to the 4 corners of the rectangle in Figure 3, ie, patient A has early-stage COVID-19 (WHO score 4) and high preexisting risk; patient B, later-stage COVID-19 (WHO score 6) and high preexisting risk; patient C, early-stage COVID-19 (WHO score 4) and low preexisting risk; and patient D, later-stage (WHO score 6) and low preexisting risk. The probabilities of these patients’ expected WHO scores on days 14 and 28 appear in the top and bottom panels of Figure 4, respectively. The recommendation for patients A, B, and C is treatment with CCP, with the most substantial benefit compared with control expected for patient A (TBI score, 0.85), followed by patients B (TBI score, 0.68) and C (TBI score, 0.36). For patient D (TBI score, 0.19), CCP treatment is not recommended given that this patient has a benefit level of B3.

External Validation
Expanded Access Program Study

Early in the pandemic in the United States, a single-arm expanded access program (EAP) sponsored by the Mayo Clinic was established to provide access to CCP for hospitalized patients with COVID-19.20,21 Mortality on day 28 was the outcome. Of those who received CCP, 8698 had all baseline characteristics for computing the TBIs. Given that no participants in a control group were available from the EAP study, an EAP sample was matched to the COMPILE control participants enrolled during the concurrent time (April to September 2020), using exact matching on categorical variables and coarsened exact matching on age, yielding 1896 patients receiving CCP and 212 patients not receiving CCP. The expanded and basic TBIs were computed for these patients, and they were stratified into the predefined B1, B2 and B3 benefit levels. The ORs for the expanded TBI were as follows: for the B1 group, 0.41 (95% CI, 0.24-0.71); B2 group, 0.71 (95% CI, 0.45-1.12); and B3 group, 1.09 (95% CI, 0.65-1.81), supporting validation of the TBI (eAppendix 26 and eTables 26-30 in the Supplement).

Emergency Use Authorization Study

Under Emergency Use Authorization (EUA), CCP was permitted outside of clinical studies22 in the United States (eAppendix 27 in the Supplement). Overall, 216 hospitalized participants (210 of whom were not receiving mechanical ventilation at time of treatment) were treated with CCP (HA Yoon, email and telephone, March 13 to April 20, 2021). The outcome was day-14 ordinal WHO scores. Patients receiving CCP were matched on age, sex, and baseline WHO status with COMPILE control participants enrolled in concurrent times (October 2020 to March 2021), resulting in a matched set of 210 patients receiving CCP and 210 control patients. The ORs of the expanded TBI satisfied the conditions for validation and were as follows: for the B1 group, 0.91 (95% CI, 0.50-1.65); B2, 1.17 (95% CI, 0.67, 2.04); and B3, 3.00 (95% CI, 1.64-5.46) (eAppendix 27 and eTables 31-34 in the Supplement).

First RCT Not in COMPILE

Data from an RCT23 external to COMPILE comparing CCP vs SOC in hospitalized adults not receiving mechanical ventilation at randomization was provided for testing the TBIs. A total of 80 patients were randomized 1:1 to CCP and SOC (eAppendix 20 in the Supplement). The study found no significant CCP treatment effect.23 The outcome was 30-day mortality. Blood type information was only available for patients randomized to CCP, and therefore, only the basic TBI was used for the validation. The ORs for the predefined benefit levels indicated validation of the TBI: for B1 group, 0.31 (95% CI, 0.02-4.41); for B2 group, 0.69 (95% CI, 0.21-2.30); and for B3 group, 0.80 (95% CI, 0.09-6.85) (eAppendix 28, eTables 35-38, and eFigure 25 in the Supplement).

Second RCT Not in COMPILE

A total of 333 patients in an RCT were randomized to 2:1 to CCP and saline.24 The outcome was a 6-point version of the ordinal WHO scale. Overall, 309 patients had all baseline covariates for computing the TBIs. Both the basic and the expanded TBIs could be tested, and both were validated. With respect to day-14 WHO 6-point ordinal outcome, the ORs for the expanded TBI were as follows: for the B1 group, 0.44 (95% CI, 0.12-1.65); B2, 0.99 (95% CI, 0.49, 2.01), and B3, 1.04 (95% CI, 0.55-1.96) (eAppendix 29, eTables 39-42, and eFigures 26 and 27 in the Supplement).

Discussion

The TBIs reported in this study consist of simple combinations of baseline patient characteristics. While continuous, the TBIs can be operationalized as discrete benefit levels for the utility of making clinical decisions about treating patients with COVID-19 with CCP. Within the B1 group, patients with TBI scores approaching 1 were expected to experience large, clinically meaningful benefits from CCP. Within the B3 group, patients with TBIs approaching 0 were expected to experience harm, while the rest of the patients in this group would likely experience no benefit. Patients in the B2 group were expected to experience modest benefit.

The proposed TBIs were validated on 4 external data sets, providing evidence of its generalizability outside COMPILE and utility in practice. While the prevalence of COMPILE individuals in the benefit level groups (ie, B1, B2, and B3) was 28%, 42% and 31% respectively, the general prevalence would depend on the composition of the COVID-19 hospitalized population in different regions of the world at different times. The prevalence of patients in COMPILE expected to benefit most from CCP (B1 group) decreased, while the prevalence of patients expected to have potential harm (B3 group) increased over the enrollment quarters (Table). This may, at least partly, explain the observed decreasing trend over time in CCP efficacy (eAppendix 5 and eFigure 2 in the Supplement).

The basic TBI can be augmented with additional pretreatment characteristics to make further refinements of the clinical recommendations. This feature can be particularly useful when data on patients’ pretreatment antibody levels or other laboratory values are available. We found that no individual characteristic alone was as effective as the TBI in characterizing HTE.

Limitations

This study has limitations. As COVID-19 continues to evolve through mutations, the TBIs developed from the COMPILE study may need to be updated to reflect these potential changes. Just as current COVID-19 vaccines may lose their effectiveness as the virus mutates,25 the optimal TBI composition might also change. Additionally, this study did not discuss the association of antibody levels in the donors’ plasma with CCP efficacy. However, the method for deriving the TBI can be extended by allowing for a continuous measure of treatment16 (eg, titer quantities in donors’ plasma).

Conclusions

The TBI presented in this study is a simple tool that provides predictions for individual patients regarding their relative benefit from treatment with vs without CCP. The proposed TBIs are implemented as an application available for desktops and mobile devices.26

Back to top
Article Information

Accepted for Publication: December 15, 2021.

Published: January 25, 2022. doi:10.1001/jamanetworkopen.2021.47375

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2022 Park H et al. JAMA Network Open.

Corresponding Author: Eva Petkova, PhD, Division of Biostatistics, Department of Population Health, New York University Grossman School of Medicine, 180 Madison Ave, Room 2-55, New York, NY 10016 (eva.petkova@nyulangone.org).

Author Contributions: Drs Park and Petkova had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Tarpey, Liu, Goldfeld, Ray, Villa, Verdun, Scibona, Burgos Pratx, Duarte, Hsue, Meyfroidt, Ortigoza, Pirofski, Troxel, Antman, Petkova.

Acquisition, analysis, or interpretation of data: Park, Tarpey, Liu, Y. Wu, D. Wu, Li, Zhang, Ganguly, Ray, Paul, Bhattacharya, Belov, Huang, Forshee, Verdun, Yoon, Agarwal, Simonovich, Burgos Pratx, Belloso, Avendaño-Solá, Bar, Duarte, Luetkemeyer, Meyfroidt, Nicola, Mukherjee, Ortigoza, Pirofski, Rijnders, Antman, Petkova.

Drafting of the manuscript: Park, Tarpey, Liu, Y. Wu, Li, Zhang, Burgos Pratx, Hsue, Pirofski, Petkova.

Critical revision of the manuscript for important intellectual content: Park, Tarpey, Goldfeld, D. Wu, Zhang, Ganguly, Ray, Paul, Bhattacharya, Belov, Huang, Villa, Forshee, Verdun, Yoon, Agarwal, Simonovich, Scibona, Burgos Pratx, Belloso, Avendaño-Solá, Bar, Duarte, Luetkemeyer, Meyfroidt, Nicola, Mukherjee, Ortigoza, Pirofski, Rijnders, Troxel, Antman, Petkova.

Statistical analysis: Park, Tarpey, Liu, Goldfeld, Y. Wu, D. Wu, Li, Zhang, Ganguly, Paul, Bhattacharya, Belov, Forshee, Burgos Pratx, Troxel, Petkova.

Obtained funding: Scibona, Hsue, Luetkemeyer, Petkova.

Administrative, technical, or material support: Park, Ganguly, Ray, Paul, Bhattacharya, Belov, Huang, Villa, Simonovich, Scibona, Burgos Pratx, Avendaño-Solá, Bar, Petkova.

Supervision: Liu, Ray, Forshee, Verdun, Scibona, Duarte, Hsue, Nicola, Pirofski, Antman, Petkova.

Conflict of Interest Disclosures: Dr Yoon reported receiving grants from the G. Harold and Leila Y. Mathers Foundation during the conduct of the study. Dr Duarte reported receiving personal fees from Amgen, Astellas, Bristol Myers Squibb, Gilead Sciences, Jazz Pharmaceuticals, Kiadis Pharma, Miltenyi Biotec, Merck Sharp and Dohme, Omeros, Pfizer, Sanofi Oncology, Sobi, and Takeda outside the submitted work. Dr Hsue reported receiving honoraria from Gilead Sciences and Merck and receiving grants from Novartis outside the submitted work. Dr Luetkemeyer reported receiving grants from the Steve and Marti Diamond Charitable Foundation during the conduct of the study and grants from Gilead Sciences, Eli Lilly and Co, and EMD Serono outside the submitted work. Dr Meyfroidt reported receiving grants from the Belgian Health Care Knowledge Center and the Research Foundation Flanders during the conduct of the study. Dr Pirofski reported receiving grants the G. Harold and Leila Y. Mathers Foundation during the conduct of the study. Dr Rijnders reported receiving grants from Erasmus MC Foundation during the conduct of the study. No other disclosures were reported.

Funding/Support: The design and conduct of the study; data collection, management, analysis, and interpretation of the data; and preparation of the manuscript for publication were supported by grant UL1TR001445 from the National Center for Advancing Translational Sciences of the National Institutes of Health (NIH). The statistical methodology and the extensions necessary for its applicability in the context of convalescent plasma use for treating COVID-19 were developed with support from grant R01MH099003 from National Institute of Mental Health of the NIH.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the US Food and Drug Administration.

Additional Contributions: Alison Bateman-House, PhD (New York University), Eric Boersma, PhD (Erasmus University), David Glidden, PhD (University of California, San Francisco), Lakshmanan Jeyaseelan, PhD (Christian Medical College), Emmanuel Lesaffre, PhD (Katholieke University of Leuven), Grigorios Papageorgiou, (Erasmus University), Aitor Perez, PhD (Pivotal CRO), Suman Pramanik, MD (Army Hospital), Aranzazu Sancho-Lopez, MD, PhD (Hospital Universitario Puerta de Hierro Majadahonda), André Siqueira, MD (University of Brasilia), John Szumowski, MD, MPH (University of California, San Francisco), Séverine Vermeire , MD, PhD (Universitait Ziekenhuis Leuven), John Younger, MD (University City Science Center), Pamela Shaw, PhD (Kaiser Permanente Washington Health Research Institute), and Geert Verbeke, PhD (Katholieke University of Leuven), served on the data safety monitoring board for the COMPILE study. Barbee Whitaker, PhD (Center for Biologics Evaluation and Research, Office of Biostatistics and Epidemiology, Food and Drug Administration), facilitated the validation of TBI in 1 of the external data sets. Michael Joyner, MD, (Mayo Clinic, Human and Integrative Physiology and Clinical Pharmacology Laboratory), conducted the study that collected 1 of the validation data sets and allowed the use of those data for validation. Rickey Carter, PhD (Mayo Clinic, Biostatistics), was the biostatistician for a study that provided 1 of the external data sets for validation of the TBI and commented on the TBI development and validation. R. Todd Ogden, PhD (Columbia University, Department of Biostatistics), and David DeMets, PhD (University of Wisconsin, Biostatistics), discussed with the study team the methodology for the TBI and its application to this study. Judith S. Hochman, MD (New York University, School of Medicine), and Corita Grudzen, MD (New York University, School of Medicine), provided administrative support for the study. None of these individuals were compensated for their time.

References
1.
Tonelli  MR, Shirts  BH.  Knowledge for precision medicine: mechanistic reasoning and methodological pluralism.   JAMA. 2017;318(17):1649-1650. doi:10.1001/jama.2017.11914PubMedGoogle ScholarCrossref
2.
Blackstone  EH.  Precision medicine versus evidence-based medicine: individual treatment effect versus average treatment effect.   Circulation. 2019;140(15):1236-1238. doi:10.1161/CIRCULATIONAHA.119.043014PubMedGoogle ScholarCrossref
3.
DeMerle  K, Angus  DC, Seymour  CW.  Precision medicine for COVID-19: phenotype anarchy or promise realized?   JAMA. 2021;325(20):2041-2042. doi:10.1001/jama.2021.5248PubMedGoogle ScholarCrossref
4.
Bzdok  D, Varoquaux  G, Steyerberg  EW.  Prediction, not association, paves the road to precision medicine.   JAMA Psychiatry. 2021;78(2):127-128. doi:10.1001/jamapsychiatry.2020.2549PubMedGoogle ScholarCrossref
5.
Zhao  Y, Zeng  D, Rush  AJ, Kosorok  MR.  Estimating individualized treatment rules using outcome weighted learning.   J Am Stat Assoc. 2012;107(449):1106-1118. doi:10.1080/01621459.2012.695674PubMedGoogle Scholar
6.
Song  R, Kosorok  M, Zeng  D, Zhao  Y, Laber  E, Yuan  M.  On sparse representation for optimal individualized treatment selection with penalized outcome weighted learning.   Stat. 2015;4(1):59-68. doi:10.1002/sta4.78PubMedGoogle ScholarCrossref
7.
Laber  EB, Zhao  YQ.  Tree-based methods for individualized treatment regimes.   Biometrika. 2015;102(3):501-514. doi:10.1093/biomet/asv028PubMedGoogle ScholarCrossref
8.
Petkova  E, Tarpey  T, Su  Z, Ogden  RT.  Generated effect modifiers (GEM’s) in randomized clinical trials.   Biostatistics. 2017;18(1):105-118. doi:10.1093/biostatistics/kxw035PubMedGoogle ScholarCrossref
9.
Liu  Y, Wang  Y, Kosorok  MR, Zhao  Y, Zeng  D.  Augmented outcome-weighted learning for estimating optimal dynamic treatment regimens.   Stat Med. 2018;37(26):3776-3788. doi:10.1002/sim.7844PubMedGoogle ScholarCrossref
10.
Park  H, Petkova  E, Tarpey  T, Ogden  RT.  A single-index model with multiple-links.   J Stat Plan Inference. 2020;205:115-128. doi:10.1016/j.jspi.2019.05.008PubMedGoogle ScholarCrossref
11.
Park  H, Petkova  E, Tarpey  T, Ogden  RT.  A constrained single-index regression for estimating interactions between a treatment and covariates.   Biometrics. 2021;77(2):506-518.PubMedGoogle ScholarCrossref
12.
Petkova  E, Antman  EM, Troxel  AB.  Pooling data from individual clinical trials in the COVID-19 era.   JAMA. 2020;324(6):543-545. doi:10.1001/jama.2020.13042PubMedGoogle ScholarCrossref
13.
Troxel  AB, Petkova  E, Goldfeld  K,  et al.  Association of convalescent plasma treatment with clinical status in patients hospitalized with COVID-19: a meta-analysis.   JAMA Netw Open. 2022;5(1):e2147331. doi:10.1001/jamanetworkopen.2021.47331Google Scholar
14.
Goldfeld  K, Wu  D, Tarpey  T,  et al.  Prospective individual patient data meta-analysis: Evaluating convalescent plasma for COVID-19.   Stat Med. 2021;40(24):5131-5151. doi:10.1002/sim.9115PubMedGoogle ScholarCrossref
15.
WHO Working Group on the Clinical Characterisation and Management of COVID-19 infection.  A minimal common outcome measure set for COVID-19 clinical research.   Lancet Infect Dis. 2020;20(8):e192-e197. doi:10.1016/S1473-3099(20)30483-7PubMedGoogle ScholarCrossref
16.
Park  H, Petkova  E, Tarpey  T, Ogden  RT.  A single-index model with a surface-link for optimizing individualized dose rules.   J Computational Graphical Stat. Published online June 21, 2021. doi:10.1080/10618600.2021.1923521Google Scholar
17.
Gray  RJ.  A class of k-sample tests for comparing the cumulative incidence of a competing risk.   Ann Stat. 1988;16(3):1141-1154. Accessed December 20, 2021. https://www.jstor.org/stable/2241622Google ScholarCrossref
18.
McCaw  ZR, Tian  L, Vassy  JL,  et al.  How to quantify and interpret treatment effects in comparative clinical studies of COVID-19.   Ann Intern Med. 2020;173(8):632-637. doi:10.7326/M20-4044PubMedGoogle ScholarCrossref
19.
Janes  H, Pepe  MS, Gu  W.  Assessing the value of risk predictions by using risk stratification tables.   Ann Intern Med. 2008;149(10):751-760. doi:10.7326/0003-4819-149-10-200811180-00009PubMedGoogle ScholarCrossref
20.
Senefeld  JW, Johnson  PW, Kunze  KL,  et al.  Program and patient characteristics for the United States Expanded Access Program to COVID-19 convalescent plasma.   medRxiv. Preprint posted online April 11, 2021. doi:10.1101/2021.04.08.21255115Google Scholar
21.
Joyner  MJ, Carter  RE, Senefeld  JW,  et al.  Convalescent plasma antibody levels and the risk of death from COVID-19.   N Engl J Med. 2021;384(11):1015-1027. doi:10.1056/NEJMoa2031893PubMedGoogle ScholarCrossref
22.
US Food and Drug Administration. FDA issues Emergency Use Authorization for convalescent plasma as potential promising COVID-19 treatment, another achievement in administration’s fight against pandemic. April 23, 2020. Accessed December 20, 2021. https://www.fda.gov/news-events/press-announcements/fda-issues-emergency-use-authorization-convalescent-plasma-potential-promising-covid-19-treatment
23.
Ray  Y, Paul  SR, Bandopadhyay  P,  et al.  Clinical and immunological benefits of convalescent plasma therapy in severe COVID-19: insights from a single center open label randomised control trial.   medRxiv. Preprint posted online November 29, 2020. doi:10.1101/2020.11.25.20237883Google Scholar
24.
Simonovich  VA, Burgos Pratx  LD, Scibona  P,  et al; PlasmAr Study Group.  A randomized trial of convalescent plasma in COVID-19 severe pneumonia.   N Engl J Med. 2021;384(7):619-629. doi:10.1056/NEJMoa2031304PubMedGoogle ScholarCrossref
25.
Madhi  SA, Baillie  V, Cutland  CL,  et al; NGS-SA Group; Wits-VIDA COVID Group.  Efficacy of the ChAdOx1 nCoV-19 COVID-19 Vaccine against the B.1.351 variant.   N Engl J Med. 2021;384(20):1885-1898. doi:10.1056/NEJMoa2102214PubMedGoogle ScholarCrossref
26.
Park H, Tarpey T, Li Y, et al. Convalescent plasma treatment benefit index calculator. January 21, 2022. http://covid-convalescentplasma-tbi-calc.org
×