Appendix CCharacteristics of Key Question 1 and 2 Studies

Ashana et al. 2021¹

*Also addressed KQ2

SOFA score

Laboratory-based Acute Physiology Score version 2 (LAPS2)

Mitigation strategy: modified versions of SOFA and LAPS2

SOFA score: Composed of organ function scores from 6 organ systems (cardiovascular, respiratory, hepatic, renal, coagulation, neurological). In this study, the renal subscore was calculated using creatinine alone, and the highest value of each subscore during the patient’s ED stay was used to calculate total score (continuous variable 0 to 24 points, with higher values representing greater illness severity).

SOFA score modifications:

1) divided score into 4 categories (<6, 6 to 8, 9 to 11, ≥ 12)

2) subtracted one-half point from renal subscore for Black patients whose raw renal subscore >0

3) eliminated renal subscore

LAPS2: 2-stage algorithm in which patients are first stratified into low- and high-mortality-risk groups, and then vital signs and laboratory values are added to the algorithm. Total score is based on risk stratum and most deranged laboratory value and ranges from 0 to 414 (continuous variable, scores > 200 uncommon).

LAPS2 modifications:

1) continuous LAPS2 divided into 8 categories

2) continuous LAPS2 divided into 4 equal categories

Modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated

EDs of academic medical centers

Patients admitted for sepsis or ARF at 27 hospitals (Kaiser Permanente Northern California and Penn Medicine) between 2013 and 2018. Patients were ≥ 18 years with sepsis at all hospitals and ARF at Penn Medicine hospitals and were admitted from the ED to an inpatient location.

Study does not report how race/ethnicity was defined (e.g., self-reported)

Patients (n): 113,158

Race/Ethnicity: 75.6% White; 24.4% Black

Mean Age (SD): 67.7 (15.2) White; 61.7 (16.6) Black

p<0.001

Sex: White 54.1% male and 45.9% female; Black 48.2% male and 51.8% female

p<0.001 for % female between groups

Rapid triage fast-track (FT) model

Providers assign an emergency severity index (ESI) score, then determine whether patients meet additional criteria for FT or main ED status.

Triage process: Nurse checks in patient, determines chief complaint, and obtains set of vital signs. Nurse applies ESI protocol to give patient a score of 1 (most acute) to 5 (least acute). Nurse then determines whether patient is appropriate for FT based on the following requirements:

1. Patient is able to sit in a recliner.
2. Patient is ambulatory and able to speak.
3. Patient’s ESI score is 3, 4, or 5 (lowest acuity).
4. Patient is determined to be not critical based on the triage determination

Patients identified for FT are seen in a separate 5-bed area of ED. Patients not eligible for FT wait until an ED bed becomes available.

FT status: patients placed in a separate ED section and can receive intravenous fluids, medications, and laboratory and radiology tests. A physician assistant, nurse practitioner, nurse, or ED technician provides care, but an ED physician can be involved if needed (e.g., case’s complexity).

Main ED status: patients wait until an ED bed is available. Care typically provided by ED physician.

Retrospec-tive matched cohort (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Tertiary care hospital (556-bed)

EHR ED encounters during a 1-year period (2/1/2019 to 1/31/2020) after full implementation of rapid triage system.

Black and White patients were exact-matched on potential confounders including presence of abnormal vital signs.

Race and ethnicity collected as separate measures in the EHR based on patient self-reporting and patients can report multiple races.

Race and ethnicity not included as an input variable.

Patients (n): 9704 with 12,330 unique encounters (5151 Black Non-Hispanic encounters; 7170 White Non-Hispanic encounters)

Race/Ethnicity: 58.2% White Non-Hispanic encounters; 41.7% Black Non-Hispanic encounters

Mean Age (SD): 37.4 (36.9) White Non-Hispanic; 36.9 (13.2) Black Non-Hispanic

Sex: 70.4% female; 29.6% male

Prostate Cancer Prevention Trial Risk Calculator 2.0 (PCPT RC)

Prostate Biopsy Collaborative Group (PBCG) RC

Retrospec-tive (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Urology clinics

The sample consisted of consecutive ambulatory patients from urology clinics at 2 privately funded and 3 publicly funded institutions who were undergoing their first prostate biopsy for an abnormal PSA level or digital rectal exam (DRE). Data obtained from a prospectively maintained dataset.

Race/ethnicity was self-reported.

Patients (n): 954

Race/Ethnicity: Black (463, 48.5%), white (355, 37.2%), Other race (136, 14.2%). The Other group included Hispanic (n = 103, 75.7%), Asian (n = 28) and Middle Eastern men (n = 5).

Median Age (IQR): Black, 61 (57, 67); white, 62 (58, 67); other, 62 (57, 67)

Sex: 100% male

Simulation study (modeling using synthetic data – source data for calculation of algorithmic scores is synthetic data and outcomes that would have resulted from using the algorithm are simulated)

1950 U.S. birth cohort aged 50–90 years

Analyses were performed on a simulated dataset of 100,000 individuals in the 1950 U.S. birth cohort, containing: (a) smoking history data generated by the CISNET Smoking History Generator based on data from the NHIS, Cancer Prevention studies I and II, and the Human Mortality Database, and (b) risk factor data generated by the LC Risk Factor Generator based on data from the NHIS, PLCO trial, U.S. Census Bureau, and NHANES.

Sensitivity analyses were performed on a similar dataset representing the 1960 U.S. birth cohort.

Race/ethnicity data generated from U.S. Census Bureau data.

1950 birth cohort:

Patients (n): 100,000

Race/Ethnicity: White (76%), Black (10%), Hispanic (8%), Asian (5%)

Age: Not reported

Sex: Not reported

A 5-level triage algorithm

ESI 1 (Immediate medical attention). Study excluded visits with an ESI score of 1.

ESI 2 (Emergency)

ESI 3 (Urgent)

ESI 4 (Nonurgent)

ESI 5 (Minor)

Retrospec-tive cohort (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Pediatric ED

EHR data from July 2015 to June 2016 for patients aged 0 to 17 years.

Study does not indicate whether race and ethnicity was self-reported. Patients were categorized as Non-Hispanic White if White/Caucasian and non-Hispanic ethnicity were reported. Patients were categorized as non-White if any other race was reported (Black/African American, Asian, American Indian/Alaska Native, Native Hawaiian/Pacific Islander, other) or if an ethnicity of Hispanic was reported.

Race and ethnicity not included as an input variable.

Patients (n): 8928 (3086 Non-Hispanic White; 5842 Non-White) with 10,815 visits (3538 Non-Hispanic White; 7277 Non-White)

Race/Ethnicity: 34.6% Non-Hispanic White; 65.4% Non-White (1.2% American Indian/Alaska Native, 14.6% Asian, 23.8% Black, 2.5% Native Hawaiian/Pacific Islander, 38.7% other, 11% more than 1 race)

Median Age (Months at Visit; IQR): 39.2 (14.1 to 88.5) Non-Hispanic White; 33.4 (13.7 to 74.2) Non-White

Sex: 46.8% female; 53.2% male

Modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated

ICUs

The SOFA score was developed using a consensus-based process. In this study, the data source is the eICU Collaborative Research Database, a cohort of patients admitted to ICUs in 208 U.S. hospitals from 2014 to 2015. Eligibility criteria: age ≥18 years; Black or white race; at least 1 SOFA variable recorded within 24 hours of ICU admission; in-hospital mortality documented. Records representing the first ICU stay during a hospitalization were included.

Study does not report how race/ethnicity was defined (e.g., self-reported).

Patients (n): 111,885 patient encounters for 95,549 unique patients

Race/Ethnicity: 16,688 encounters with Black patients (14.9%) and 95,197 encounters with White patients (85.1%)

Mean (SD) age: 63.3 (16.9) years

Sex: 51,464 encounters with women (46.0%)

Obermeyer et al. 2019⁷

*Also addressedKQ2

Features of raw insurance claims data from the previous year, including age, sex, insurance type, diagnosis and procedure codes, medications, and detailed costs.

Race is not an input variable in the original algorithm or in the 3 new algorithms.

Retrospective cohort study (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Hospital

Patients (n): 49,618

Race/Ethnicity: 87.7% White, 12.3% Black

Mean Age: 51.3 (White), 48.6 (Black)

Sex: 62% female (White), 69% female (Black)

USPSTF 2013: NLST-like eligibility criteria include age 55 to 80 years, ≥ 30 pack-year cigarette smoking history, and having quit smoking within the past 15 years.

PLCOm2012: age, highest level of education obtained, BMI, COPD, personal history of cancer, family history of lung cancer, race and ethnicity, smoking status (former or current), average number of cigarettes smoked per day, duration smoked, years of quitting smoking

Retrospective study (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Urban academic medical center (13 federally qualified health centers)

Lung cancer cohort at the University of Illinois Hospital and Health Sciences System between 2010 and 2019. Data collected up until March 15, 2020.

Study does not report how race/ethnicity was defined (e.g., self-reported)

Patients (n): 883

Race/Ethnicity: 56.3% African American; 29.2% White; 7.8% Hispanic; 2.7% Asian; 4.0% Other or missing

Mean Age (SD): 64.8 (9.4)

Sex: 55.8% male; 44.2% female

Prostate cancer risk prediction

Kaiser Permanente Prostate Cancer Risk Calculator (KPPC RC) (range: 0% to 100%)

Version A: age, race (patient-reported), BMI, family history of prostate cancer, number of prior biopsies, PSA level, DRE result

Version B: Version A variables plus prostate volume

(The study also examined a version that did not include DRE result or prostate volume, but did not report results by race or ethnicity for that version.)

Retrospective

Large integrated health care system (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Validation (this study): Kaiser Permanente Northern California (KPNC)

All men with no prior diagnosis of prostate cancer who underwent prostate biopsy at any of 21 KPNC urology departments between 12/2017 and 8/2019 for either an abnormal DRE and/or an elevated PSA and had complete data on all analysis variables. Prospective data collection methods used to capture biopsies during this time.

Study does not report how race/ethnicity was identified.

Patients (n): 4178 (5353 men underwent biopsy; 4178 had complete data)

Race/Ethnicity: 56.7% Caucasian, 11.3% African American, 16.2% Asian/Pacific Islander, 13.5% Hispanic, 2.3% other

Median Age (IQR): 63 (57–67)

Sex: 100% male

CSC scoring system: aggregate score outlined by Commonwealth of Massachusetts for application in individual hospitals. Score based on points derived from SOFA score and a chronic severity of illness score based on comorbidities or a life expectancy score based on physician assessment. SOFA converted into a 4-point scale: 1 for SOFA < 6, 2 for SOFA 6 to 9, 3 for SOFA 10 to 12, and 4 for SOFA>12. Comorbidities based on a 3-level system: 0 points no significant comorbidities, 2 points major comorbid conditions with substantial impact on long-term survival, 4 points severely life-limiting conditions prior to acute illness. Life expectancy based on a 3-level score: 0 points death not likely in 5 years, 2 points death likely within 5 years, and 4 points death likely within 1 year. Points totaled to create raw ordinal priority score from 1 to 8. Highest scores 1 to 2, intermediate 3 to 5, and lowest 6 to 8. Highest-scores first to receive scarce critical care resources, then intermediate scores, then lowest scores.

Simulation: Simulation of mortality outcomes using CSOC score vs random lottery in a subset of patients receiving ventilation. Created a scenario of scarcity requiring allocation of ventilators using 2 state-recommended cutoff scores of ≤ 2 (highest-priority category patients receive ventilator) and ≤ 5 (both highest and intermediate-priority category patients receive ventilator). Authors ran 10,000 trials randomly assigning individuals to receive a ventilator.

Retrospective cohort (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

ICUs of 6 Boston-area tertiary and community hospitals in Beth Israel and Lahey Hospital systems

Patients (n): 498 (79 Black, 298 White, 11 Asian, 46 Other, 64 Unknown)

Race/Ethnicity (%): 15.8% Black, 59.8% White, 2.2% Asian, 9.2% Other, 12.8% Unknown

Median Age (IQR), years: 67 (56 to 75)

Black 68 (59 to 75), White 69 (57 to 76), Asian 62 (59 to 72), Other 63 (52 to 73), Unknown 59 (50 to 69)

Sex: 38,4% female; 61.6% male

Black (32.9% female, 67,1% male), White (39.3% female, 60.7% male), Asian (36.4% female, 63.6% male), Other (32.6% female, 67.4% male), Unknown (45.3% female, 54.7% male)

Subgroup (n): 244 (16.8% Black, 49.2% White, 2.9% Asian, 10.7% other, 20.5% unknown)

Race/Ethnicity (%): 16.8% Black, 49.1% White, 2.8% Asian, 10.6% other, 20.4% unknown)

APACHE IVa generates a risk score for hospital, ICU mortality, and length of stay.

OASIS predicts hospital mortality and ICU mortality of critically ill patients

SOFA characterizes severity state in sepsis but has been used to predict patient outcomes

APACHE IVa: 142 patient variables including 116 admission categories and 17 acute physiologic parameters (65.9% of score and includes age, chronic health condition, underlying diagnosis, ventilation status).

OASIS: 10 variables collected in first 24 hours of ICU stay (heart rate, mean arterial pressure, temperature, respiratory rate, urine output, pre-ICU admission length of stay, GCS, age, being placed on a mechanical ventilator at any point during day 1 and admission following elective surgery).

SOFA: composed of organ function scores from 6 organ systems (cardiovascular, respiratory, hepatic, renal, coagulation, neurological). SOFA categories based on proposed categories for COVID-19 ventilator allocation.

Modeling study using real-world data to determine effect of illness severity scores on CSC allocation (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

ICU admissions

eICU-Collaborative Research Database (eICU-CRD) includes > 200,000 discharge episodes across 335 ICUs at 208 hospitals between 2014 and 2015. Data available includes age, sex, ethnicity, vital signs, diagnoses, laboratory measurements, clinical history, problem lists, APACHE IVa scores, and treatments.

Medical Information Mart for Intensive Care-III database consists of >60,000 ICU admissions to Beth Israel Deaconess Medical Center between 2001 and 2012 and includes OASIS as a mortality prediction model.

eICU-CRD

Patients (n): 122,919

Race/Ethnicity: 81.9% White; 12.4% African American; 4.1% Hispanic; 1.5% Asian

Median Age (IQR): 64 (52 to 75)

Sex: 54% male; 46% female

MIMIC-III

Patients (n): 43,823

Race/Ethnicity: 82.14% White; 11.07% African American; 4.07% Hispanic; 2.71% Asian

Median Age (IQR): 64.5 (52 to 76)

Sex: 57% male; 43% female

Note: race/ethnicity is based on definitions within each database. Information is typically entered by an administrator. Patients are either asked which group they identify with or the group is entered based on available records.

HEART Pathway risk assessment is based on the HEAR score (History, ECG, Age, and Risk factor) and 0- and 3-hour troponin measures.

HEAR score ≤ 3 without elevated troponin is classified as low-risk and recommended for discharge without objective cardiac testing. HEAR score ≤ 3 with elevated troponin leads to a cardiology consult and admission and/or further observation or testing.

HEAR score ≥ 4 with elevated troponin, known coronary artery disease, or ischemic ECG is classified as non-low risk and designated for further testing. HEAR score ≥ 4 without elevated troponin leads to observation/admission and/or cardiology consult or testing.

*Authors focused on low-risk (≤ 3) and non-low risk (≥ 4) groups for analysis.

Preplanned subgroup analysis of a prospective pre-post study (pre-post study).

3 EDs in North Carolina (large urban academic medical center, rural medical center, small community hospital).

EHR (Clarity-EPIC systems) index encounter and claims data.

Race/ethnicity was self-reported.

Patients (n): 3713 pre-implementation; 4,761 post-implementation

Race/Ethnicity Pre-implementation: 66.9% (2,484) White; 28.3% (1,052) Black or African American; 0.6% (21) Asian, 0.2% (9) American Indian, 0.03% (1) Hawaiian or Pacific Islander, 3.9% Other (145), 0.03% (1) Refused to provide information or Unknown

Race/Ethnicity Post-implementation: 65.2% (3,106) White, 28.8% (1,371) Black, 0.6% (27) Asian, 0.3% (16) American Indian, 0.04% (2) Hawaiian or Pacific Islander, 4.9% (234) Other, 0.1% (5) Refused to provide information or Unknown.

Median Age: 54 years

Sex: 46.4% male; 53.6% female

Thompson et al. 2021¹³

*Also addressed KQ2

Retrospective cohort study (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Hospital and tertiary care academic center

Development dataset: adult hospital encounters from EHR between 2007 and 2017 at a U.S. hospital and tertiary academic center. Opioid-related hospitalizations were oversampled. “The final dataset …consisted of 367 manually labeled cases, age- and sex-matched with controls that had no indications of opioid misuse.” External validation dataset: EHR at a different tertiary care academic center. Dataset included “all unplanned adult inpatient encounters …who were screened between October 23, 2017, and December 31, 2019 (n = 53,974).”

Appears race/ethnicity was self-reported.

The analyses reported in this article were carried out on the external validation dataset.

Patients (n): 53,794

Race/Ethnicity: White (n = 23,345), Black (n = 17,541), Hispanic/Latinx (n = 9252), Other (n = 3836)

Age: not reported

Sex: not reported

In 2005, LAS became the main method for determining allocation of deceased donor lungs for transplantation in the United States.

The LAS is a numerical score based on survival models that estimate likelihood of survival both while on the wait list and post-transplant; thus, it reflects the net benefit of transplantation.

Retrospective pre/post-implementation study (pre-post study)

U.S. health care system

The study population consisted of all White and Black non-Hispanic adults who were listed for lung transplantation during 2 time periods: pre-LAS (January 1, 2000–May 3, 2005) and LAS (May 4, 2005–September 4, 2010).

Race/ethnicity was self-reported.

Patients (n): 8765 (pre-LAS), 8806 (LAS).

Race/Ethnicity: White (89.9%), Black (10.1%)

Mean (SD) Age: Pre-LAS: White, 49.3 (12.6); Black, 47.2 (9.6)

LAS: White, 54.0 (13.0); Black, 50.4 (10.5)

Sex: Pre-LAS: 51.3% female

LAS: 45.5% female

Patients (n): 41,544

Race/Ethnicity:

White-Non-Hispanic (n=36,787); Black Non-Hispanic (n=66);

Hispanic (n=786); Other Non-Hispanic (n=1905)

No comparator. Compared outcomes by racial and ethnic groups

*Calculator outputs are used to guide clinical guideline-based care

CHA₂DS₂-VASc (start at 0): age 65 to 74 (+1) or > 75 (+2), female (+1), CHF history (+1), HTN history (+1), stroke / TIA / thromboembolism history (+2), vascular disease history (+1), diabetes history (+1).

The algorithm informs the American College of Cardiology (ACC)/American Heart Association (AHA) atrial fibrillation treatment guideline.

2014 ACC/AHA guideline recommendation: do not recommend antithrombotic therapy for male patients with a score of 0 or female patients with a score of 1.

2020 ACC/AHA guideline recommendation: recommend antithrombotic therapy for male patients with a score ≥ 2 and female patients with a score ≥ 3. Consider antithrombotic therapy for male patients with a score of 1 and female patients with a score of 2.

Retrospective cohort (modeling using real-world data – source data for calculation of algorithmic scores is real world data and outcomes that would have resulted from using the algorithm are simulated)

Stanford Health Care and Lucile Packard Children’s Hospital

Stanford Medicine Research Data Repository (STARR). STARR was linked with the Social Security Administration’s Death Master File to determine out-of-hospital deaths.

Race and ethnicity self-reported. Study used the 5 U.S. Census Bureau categories and used Hispanic as a dedicated ethnicity.

Race and ethnicity are not included as input variables in MELD, CHA₂DS₂-VASc, or sPESI.

CHA₂DS₂-VASc

Patients (n): 233,129

Race/Ethnicity: Asian 15% (n=33,927), Black 3% (n=7323), White 76% (n=176,278), Hispanic 6% (n=13,578), Other 1% (n=2,023)

Median Age (25^th to 75^th percentiles): 77 years (71 to 83)

Sex: 56% male (n=129,621), 44% female (n=103,508)

Kidney allocation

KAS was developed to improve equity related to dialysis time and to patients with high panel reactive antibody. Specific changes to the system include a change in the calculation of waiting time and prioritization of the most sensitized patients. Waiting time starts at dialysis start instead of at the time of waitlist.

KAS uses Kidney Donor Profile Index (KDPI) and Expected Post Transplant Survival (EPTS) score for longevity matching between donors and recipients.

KDPI (donor variables): age, height, weight, ethnicity, history of hypertension, history of diabetes, cause of death, serum creatinine, hepatitis C Virus status, donation after Circulatory Death status (range: 0% to 100%). Options for ethnicity variable: American Indian or Alaska Native, Asian, Black or African American, Hispanic/Latino, Native Hawaiian or Other Pacific Islander, White, or Multi Racial. EPTS score: age, time on dialysis, current diabetes status, and if candidate had a previous solid organ transplant (range: 0% to 100%).

KAS also incorporates a points system to increase priority for patients with high panel reactive antibody (i.e., patients less likely to find a compatible donor), includes pre-registration dialysis time as part of a candidate’s waiting time, provides increased access for candidates with blood type B, uses KDPI scores to inform pediatric priority, and eliminates the payback system (i.e., if an organ was received from another organization, the receiving organization had to pay back an organ to the national pool). *Information from the Organ Procurement & Transplantation Network New KAS FAQs.

Retrospective cohort study (pre-post study)

U.S. medical centers

New patients on dialysis and existing patients on dialysis with end-stage renal disease (ESRD) from the United States Renal Data System (USRDS).

Pre-KAS group: beginning dialysis between 1/1/2005 and 12/03/2014

Post-KAS: beginning dialysis between 12/4/2014 and 12/31/2015

The United Network for Organ Sharing system used to collect information about active and inactive status of newly waitlisted patients from 2005 to 2015.

Study does not report how race/ethnicity was defined (e.g., self-reported)

Pre-KAS (incident patients)

Patients (n): 1,120,655

Race/Ethnicity: 52.1% White; 26.5% Black; 13.7% Hispanic; 4.2% Asian

Age group, N (%):

18 to 39: 7.5%

40 to 49: 10.7%

50 to 59: 19.9%

60 to 69: 24.9%

≥ 70: 47%

Sex: 56.8% male; 43.2% female

Post-KAS (incident patients)

Patients (n): 132,445

Race/Ethnicity: 51% White; 25.1% Black; 13.7% Hispanic; 4.8% Asian

Age group, N (%):

18 to 39: 7.3%

40 to 49: 10.4%

50 to 59: 19.7%

60 to 69: 27.2%

≥ 70: (35.4%)

Sex: 58.2% male; 41.8% female

Prevalent Dialysis Cohort

*Patients eligible for first time waitlisting anytime during period (1/1/2005 to 12/31/2015). Baseline characteristics NR for race/ethnicity, age, and sex.

Patients (n): 1,556,954

eGFR

(CKD-EPI)

Patients (n): 56,485

Race/Ethnicity: 87% White; 3.9% African American; 2.3% Asian; 1.0% Hispanic; 0.1% Native American; 6.1% Other

Median Age: White 77; African American 73; Asian 77; 74 Hispanic; Native American 73; Other 76

Sex: 56.5% female; 43.5% male

Ashana et al. 2021¹

*Also addressed KQ1

Illness severity prediction models

SOFA score

Laboratory-based Acute Physiology Score version 2 (LAPS2)

Patients admitted for sepsis or ARF at 27 hospitals (Kaiser Permanente Northern California and Penn Medicine) between 2013 and 2018 Patients were ≥ 18 years with sepsis at all hospitals and ARF at Penn Medicine hospitals.

Study does not report how race/ethnicity was defined (e.g., self-reported)

Patients (n): 113,158

Race/Ethnicity: 75.6% White; 24.4% Black

Mean Age (SD): 67.7 (15.2) White; 61.7 (16.6) Black

p<0.001

Sex: White 54.1% male and 45.9% female; Black 48.2% male and 51.8% female

p<0.001 for % female between groups

Patients (n): 2652

Race/Ethnicity: 20% Black

Mean age (SD): 65 (8.4) White; 58 (8.9) Black

Sex: 46% female

eGFR

(CKD-EPI)

Patients (n): 3873

Race/Ethnicity: 42.1% Black; 57.9%

Mean age (SD): 57.8 (10.9)

Sex: Black 51.2% female; Non-Black 40.9% female

eGFR

(CKD-EPI)

Patients (n): 340

Race/Ethnicity: 100% Black

Median age: 57

Sex: 49% female

GFR estimated from metabolic panel without adjustment for race or use of creatinine, as follows:

eGFR=exp(3.04584 − 0.450817*ln(Acetyl-L-Threonine) − 0.214876*ln(Beta-pseudouridine) − 0.253004*ln(Myo-inositol) + 0.2265693*ln(Tryptophan))

Patients (n): 265

Race/Ethnicity: 46% Black

Mean age (SD): 71 years (9)

Sex: 47% female

eGFR

(CKD-EPI 2021 and 2009; MDRD 2006)

Data drawn from NHANES, 2001–2018

The NHANES data were extrapolated to the US population at large.

Patients (n): 44,360 after removing patients with age <18 years or censored age; pregnant patients; and patients without serum creatinine reported.

Race/Ethnicity: 21.5% non-Hispanic Black; 41.9% non-Hispanic White; 26.6% Mexican American or Other Hispanic; 9.98% Other Race – Including Multi-Racial

Mean age (IQR): 45 (26)

Sex: 50.7% female

Patients (n): 66,987

9,945 from a Black donor

Patients (n): 6,604

Race/Ethnicity: 40% Black

Mean age: 64

Sex: 50% female

eGFR

(CKD-EPI)

NHANES cohort

Patients (n): 227,613,357

Race/Ethnicity: 11% Black

Mean age (SD): 47.4 (17.5)

Sex: 52% female

VA cohort

Patients (n): 4,477,675

Race/Ethnicity: 17% Black

Mean age (SD): 62.9 (15.8)

Sex: 8% female

Spirometry reference equations

Global Lung Function Initiative (GLI) reference equation

Patients (n): 3344

Race/Ethnicity: 36% White; 25% Black; 23% Hispanic; 17% Asian

Mean age (SD): 65.3 (9.6)

Sex: 50% female

Patients (n): 12,556

Race/Ethnicity: 10% Black

Sex: 56% female

1)

Group recalibrated algorithm model

2)

Equalized odds algorithm model

Patients (n): 25,619

Race/Ethnicity and Sex:

17.3% Black women; 11.4%

Black men; 37.8% Non-Black women; 33.4% Non-Black men

Mean age: 56.5

ASCVD event incidence: 7.54%

Added 10 biomarkers to standard CVD risk models.

10 biomarkers: adiposity (adiponectin and leptin), neurohormonal activation (aldosterone, B-type natriuretic peptide [BNP], and cortisol), inflammation (high-sensitivity C-reactive protein [hs-CRP]), endothelial function (endothelin and homocysteine), glycemic control (glycated hemoglobin), and insulin resistance (homeostasis model assessment of insulin)

Patients (n): 3689

Race/Ethnicity: 100% Black

Mean age (SD): 53 (11)

Sex: 65% female

eGFR

(CKD-EPI)

Patients (n): 62,011

Race/Ethnicity: 33.5% Black; 66.5% Non-Black

Mean age: 63

Sex: 53% female

Patients (n): 3614

Race/Ethnicity: 9.4% Black or Hispanic

Mean age: 73

Sex: 56% female

eGFR

(MDRD)

Patients (n): 567

Race/Ethnicity: 67.7% White; 32.3% Black

Mean age of patients listed preemptively: White 54.7; Black 52.1

Mean age of patients listed on dialysis: White 53.1; Black 51.4

Sex of patients listed preemptively: White 66.3% male, 33.7% female; Black 54.3% male, 45.7% female

Sex of patients listed on dialysis: White 66.2% male, 33.8% female; Black 75.5% male, 24.5% female

eGFR

(MDRD)

Test cohort

Patients (n): 947,091 procedures

Race/Ethnicity: 7.9% Black

Mean age: 64.8

Sex: 32.8% female

Validation cohort

Patients (n): 3,063,853 procedures

Race/Ethnicity: 8.5% Black

Mean age: 65.3

Sex: 31.9% female

Patients (n): 2245

Race/Ethnicity: 24% Black

Mean age (SD): 52.8 (12.8)

Sex: 29% female

1)

Removed race

2)

Replace race with cystatin C

Patients (n): 4050

Race/Ethnicity: 100% Black

Patients (n): 622 kidney donors and 1,149 recipients

Race/Ethnicity: All donors were Black

Patients(n): 460,417

Race/Ethnicity: 7% Black

Mean age: 79

Patients (n): 1015

Race/Ethnicity: 27% Black

Median age: 59 (genotype group); 57 (standard group)

Sex: 49% female

Examine whether USPSTF-2020 guidelines reduced racial and ethnic disparities compared with USPSTF-2013 guidelines and whether using an individualized prediction model for life-years gained from screening could reduce racial and ethnic disparities by identifying high-benefit individuals ineligible under USPSTF-2020 guidelines.

*Only USPSTF-2020 and USPSTF-2020 plus LYFS-CT included for this report.

USPSTF-2020

Patients (n): 14,508,450

Race/Ethnicity: 85.4% White; 8.0% African American; 1.9% Asian American; 4.7% Hispanic American

Age range: 50 to 79

Sex: 43.9% female

2020 plus LYFS-CT

Patients (n): 17,977,980

Race/Ethnicity: 82.7% White; 10.6% African American; 1.9% Asian American; 4.8% Hispanic American

Age range: 50 to 79

Sex: 44.2% female

Patients (n): 1357

Race/Ethnicity: 44% African-American

Mean age (SD): 61 (15.8)

Sex: 49% female

Patients (n): 2298

Race/Ethnicity: 63.1% White, 35.2% Black, 1.6% Other, 1.7% Hispanic

Sex: 48.8% female

eGFR

(MDRD-4, MDRD-6, CKD-EPI)

Patients (n): 72,267 patients with cirrhosis

Race/Ethnicity: 57.8% White; 19.7% Black; 7.2% Hispanic; 1.3% Asian; 13.9% Other

Median Age (IQR): 61 (57 to 66)

Sex: 2.7% female

*study reports data for Black patients only

eGFR

(CKD-EPI)

Patients (n): 25,512

Race/Ethnicity: 2.5% Black

Age (SD):* Black patients 50.6 (13.9); Non-Black patients 55.5 (14.0)

*Study does not report whether age is mean or median

eGFR

(CKD-EPI, Deindexed CKD-EPI, CG CrCl)

Patients (n): 210

Race/Ethnicity: 84.3% Black; 15.7% White

Median Age: Black 60.2; White 64.9

Median age higher among Whites compared to Blacks, p=0.001)

Sex: 42.9% female

eGFR

(CKD-EPI 2021 and CKD-EPI 2009)

Patients (n): 69,135

Race/Ethnicity: 45% Black; 40% White, 11.6% Hispanic

Mean age (SD): Black 44.4 (11.6); White 45.4 (11.3)

Sex: Black 21.8% female; White 8.2% female

Obermeyer et al. 2019⁷

*Also addressed KQ1

Patients (n): 49,618

Race/Ethnicity: 87.7% white, 12.3% Black

Mean Age: 51.3 (white), 48.6 (Black)

Sex: 62% female (white), 69% female (Black)

eGFR

(MDRD-4, CKD-EPI)

Patients (n): 7937 patients eligible for liver transplantation

Race/Ethnicity: 100% Black

Median Age (IQR): No waitlist CKD 55 (46 to 61); Waitlist CKD 58 (53 to 62)

*difference in age between groups, p<0.001

Sex: No-waitlist CKD 39.6% female; Waitlist CKD 47.8% female

*difference in sex between groups, p<0.001

Recalibrated through reweighing key groups during model training

1)

Removed race

Added a regularization term that adjusts the algorithm to limit the effect of race-based variables

Patients (n): 532,802

Race/Ethnicity: 38% Black

Mean age (SD): 26 (5.4)

Sex: 100% female

eGFR

(CKD-EPI, MDRD)

Patients (n): 459

Race/Ethnicity: 100% Black

Median age (SD): 60 (8)

Sex: 41% female

Stage 3 or 4 cancer: 29.4%

Patients (n): 241,760

Race/Ethnicity: 69% White; 10% Asian; 9% Black

Median age: 53

Sex: 51% female

Patients (n): 294 patients with hepatitis-C

Race/Ethnicity: 100% Black

Thompson et al. 2021¹³

*Also addressed KQ1

Development dataset consisted of adult hospital encounters from the EHR between 2007 and 2017 at a U.S. hospital and tertiary academic center. Opioid-related hospitalizations were oversampled. “The final dataset …consisted of 367 manually labeled cases, age- and sex-matched with controls that had no indications of opioid misuse.”

The external validation dataset came from EHR at a different tertiary care academic center. The dataset included “all unplanned adult inpatient encounters …who were screened between October 23, 2017, and December 31, 2019 (n = 53,974).”

Appears that race/ethnicity was self-reported.

Patients (n): 53,794

Race/Ethnicity: White 23,345; Black 17,541; Hispanic/Latinx 9,252); Other 3,836

Patients (n): 1231

Race/Ethnicity: 37% Black

Mean age (SD): 53 (7)

Sex: 59% female

Patients (n): 2401

Race/Ethnicity: 100% Black

Atherosclerosis Risk in Communities (ARIC) study;

Multi-ethnic Study of Atherosclerosis (MESA);

United Kingdom Biobank

Patients (n): 18,961

Race/Ethnicity: 31% African ancestry

Sex: 55% female

Added data from Jackson Heart Study and MESA to better reflect racial and ethnic populations;

Adjusted statistical methods to reduce model overfitting by using elastic net regularization; removed race-based subgroups

Derived from 6 cohort studies:

Atherosclerosis Risk in Communities Study (ARIC), Cardiovascular Health Study, Coronary Artery Risk Development in Young Adults Study, Framingham Health Study offspring cohort, Jackson Heart Study (JHS), MESA

Patients (n): 26,689

Race/Ethnicity: 29% Black

Mean age: 57

Sex: 56% female

Patients (n): 327

Race/Ethnicity: 100% Black

Mean age (SD): 62 (14.2)

Sex: 60% female

90% had diagnosis of hypertension

Patients (n): 1658

Race/Ethnicity: 100% Black

Mean Age (SD): 58 (11)

Sex: 51% female