Differentiation of Prior SARS-CoV-2 Infection and Postacute Sequelae by Standard Clinical Laboratory Measurements in the RECOVER Cohort

Abstract

Background

There are currently no validated clinical biomarkers of post-acute sequelae of SARS-CoV-2 infection (PASC).

Objective

Investigate clinical laboratory markers of SARS-CoV-2 and PASC.

Design

Propensity score-weighted linear regression models were fitted to evaluate differences in mean laboratory measures by prior infection and PASC index (≥12 vs 0).

Setting

83 enrolling sites

Participants

RECOVER Adult Cohort participants with or without SARS-CoV-2 infection, with a study visit and laboratory measures 6 months after index date (or at enrollment if >6 months). Participants were excluded if the 6-month visit occurred within 30 days of reinfection.

Measurements

Participants completed questionnaires and standard clinical laboratory tests.

Results

Of 10,094 participants, 8,746 had prior SARS-CoV-2 infection, 1,348 were uninfected; 1880 had a PASC index ≥12, and 3351 had a PASC index of zero. After propensity score adjustment, participants with prior infection had lower mean platelets (265.9 x10⁹/L, 95% CI = [264.5, 267.4]) compared to participants without known prior infection (275.2 x10⁹/L [268.5, 282.0]), and higher mean HbA1c (5.58% [5.56, 5.60] vs. 5.46% [5.40, 5.51]) and urinary albumin/creatinine ratio (81.9 mg/g [67.5, 96.2] vs. 43.0 mg/g [25.4, 60.6]), though differences were of modest clinical significance. The HbA1c difference was attenuated after excluding participants with pre-existing diabetes. Among participants with prior infection, we found no meaningful differences in mean laboratory values between those with PASC index ≥12 versus PASC index of zero.

Limitations

We could not determine whether laboratory differences represent consequences of or risk factors for acquisition of SARS-CoV-2.

Conclusions

Overall, we found no evidence that any of the 25 routine clinical laboratory values assessed in this study could serve as a clinically useful biomarker of PASC.

Primary Funding Source

National Institutes of Health

Introduction

Post-acute sequelae of SARS-CoV-2 infection (PASC), also known as Long COVID, has been reported in millions of people worldwide and is a major public health burden (1-3). PASC is generally accepted as an umbrella term encompassing a wide spectrum of symptoms and health conditions that persist after acute COVID-19 infection, resulting in a major impact on quality of life (4-12). While potential models of pathogenesis have been postulated to include immune dysregulation, viral persistence, organ injury, endothelial dysfunction, and gut dysbiosis (13-17), there are currently no validated clinical biomarkers of PASC.

As part of the National Institutes of Health's Researching COVID to Enhance Recovery (RECOVER) Initiative (https://recovercovid.org/), we previously examined prospectively collected data from nearly 10,000 individuals in the RECOVER-Adult cohort with and without SARS-CoV-2 infection and identified 12 symptoms that most distinguish those with prior infection from those uninfected (8) (Table S1). We developed a PASC index based on these 12 symptoms, with 23% of the cohort with prior SARS-CoV-2-infection meeting this research threshold. We further identified multiple clusters or subphenotypes of PASC. This framework does not encompass all individuals experiencing PASC, but permits exploration of clinical laboratory features among those meeting the PASC threshold.

Routine clinical laboratory tests that accurately distinguish individuals with PASC from those without PASC might be useful in the diagnosis, prognosis, prevention, and treatment of PASC or PASC subtypes. Laboratory tests might also identify those who have persistent organ damage but minimal to no symptoms. Studies have found potential biomarkers associated with PASC using mostly research-focused assays, but results have been inconsistent, perhaps due to different PASC study definitions, use of only selected biomarkers, choice of comparison groups if any (PASC recovered or healthy controls), duration of symptoms, types of symptoms or phenotypes, and patient population features (e.g., sex, age, race, vaccination status, comorbidities, severity of initial infection, etc.) (18-20). Systematic reviews have identified candidate biomarker categories including inflammatory (e.g., higher levels of C-reactive protein [CRP], leukocytes), coagulation (e.g., international normalized ratio [INR]), and hematologic (e.g., lower levels of hemoglobin) which likely reflect multifaceted pathophysiology and phenotypes of PASC (18, 19, 21). Autoimmune (22, 23), hormonal (24), viral (25, 26), and other (20) biomarkers associated with PASC phenotypes have also been described, though many studies have been limited by small sample sizes, limited follow-up duration, and/or lack of appropriate controls. Early small cohort studies failed to find routine clinical biomarkers (27). There is a paucity of large studies examining the utility of standardized laboratory tests obtained in routine clinical care.

Accordingly, we conducted a study to determine whether SARS-CoV-2 infection led to persistent changes in common clinical laboratory tests, regardless of symptoms, in individuals with prior infection compared to those without prior infection. If so, laboratory studies could be used to augment symptom-based definitions of PASC. Second, we aimed to determine whether those with PASC have persistent laboratory changes compared to those unlikely to have PASC. If so, we might be able to identify specific physiologic abnormalities driving symptomatic PASC. Therefore, we analyzed 25 routinely used and standardized laboratory tests, selected based on availability across different institutions, prior literature, and clinical experience (28). These tests were prospectively measured in Clinical Laboratory Improvement Amendments (CLIA)-certified laboratories in samples from 10,094 RECOVER Adult participants, representing a diverse cohort from across the U.S. We compared results 1) between participants with and without prior SARS-CoV-2 infection at ≥6 months after index infection; 2) between participants with and without PASC, as defined by an index of ≥12 or zero, respectively; and 3) between participants with each of four previously characterized PASC symptom phenotypes and those unlikely to have PASC.

Methods

Study design.

The RECOVER-Adult study design has previously been described (28). Briefly, participants were recruited at 83 enrolling sites across 33 U.S. states plus Washington, DC and Puerto Rico. Adults aged 18 years or older were eligible to enroll regardless of prior infection with SARS-CoV-2. All participants completed a baseline set of surveys, a minimal physical examination, and standard laboratory test collection at enrollment. Participants were followed prospectively with survey collection every 3 months and laboratory collection at enrollment, 6-, 12-, 24-, 36-, and 48-months post-infection or negative test date (index date).

Participants.

Participants were enrolled between 0 days and 3 years after index date and followed a visit schedule based on time from index. All participants had laboratory tests at the 6-month visit or at the enrollment visit (if enrollment was more than 6 months after index), Table S2; however, after this 6-month or enrollment visit, laboratory tests were only repeated if the prior result was abnormal. Analysis was therefore based on the 6-month visit (or the enrollment visit if more than 6 months from index) and subsequent visits were not considered, to minimize bias. Index date was defined as date of first infection (suspected, probable, or confirmed SARS-CoV-2 infection as defined by World Health Organization criteria or positive SARS-CoV-2 infection-specific antibody testing), or date of a negative SARS-CoV-2 test for participants who were never infected. Participants were excluded if the 6-month visit was within 30 days of reinfection. Participants who did not start the protocol or had no symptom survey data were excluded. Participants who had no prior SARS-CoV-2 at entry but reported infection within 6 months were reclassified as having SARS-CoV-2 infection, with a new index date defined as date of infection. Uninfected participants reporting an on-study infection >6 months after index were considered uninfected and the 6-month visit was used in the analysis. Participants pregnant at the time of infection were included in the analysis.

Exposures and outcomes.

The primary outcome was laboratory measurement at 6 months after index date, or at enrollment if >6 months. Two primary exposures included history of SARS-CoV-2 infection and PASC classification at the same time point as laboratory measurements.

All laboratory measures were completed locally at each site’s CLIA-certified clinical laboratory. Laboratory measures were drawn as part of the study visit if not obtained through routine clinical care within 30 days of index date or 90 days prior to the 6 month follow-up. Laboratory studies collected were complete blood count with differential, complete metabolic panel, INR, d-dimer, lipid panel, 25-hydroxy vitamin D, thyroid stimulating hormone (TSH), free T4, hemoglobin A1c (HbA1c), high sensitivity (hs)-CRP, cystatin C, N-terminal pro b-type natriuretic peptide (NT-proBNP), troponin, urinalysis, and urine albumin to creatinine ratio (uACR). These tests were selected based on their routine availability and standardized use across CLIA-certified laboratories, prior literature, and clinical expertise of RECOVER investigators.

PASC was defined based on our prior work (8). Briefly, we identified 12 symptoms that best distinguished between individuals with and without prior infection, assigned each one points based on degree of difference, and summed all points corresponding to the symptoms present in an individual to construct an index (Table S1). A total of 12 or more was considered the optimal threshold above which participants were likely to have PASC. Comparisons were made to participants with a PASC index of zero. A PASC index of zero does not necessarily mean that the participant had no symptoms or no PASC; however, participants with an index of zero would be unlikely to have PASC. PASC sub-phenotypes were previously defined as cluster 1 representing high frequency of impairments in smell and taste; cluster 2 as high frequency of post-exertional malaise (PEM), defined as symptoms worse after even minor physical or mental effort, and fatigue; cluster 3 as high frequency of brain fog, PEM, and fatigue; and cluster 4 as high frequency of fatigue, PEM, dizziness, brain fog, gastrointestinal symptoms, and palpitations(8).

Statistical analysis.

Demographic characteristics of the cohort were reported overall, by history of prior SARS-CoV-2 infection, and by PASC status. In the primary analysis, participants with and without a history of SARS-CoV-2 infection were compared to determine whether SARS-CoV-2 infection led to persistent laboratory abnormalities, regardless of symptoms. In the secondary analysis, participants with prior infection with PASC index ≥12 were compared to participants with a PASC index of zero. In all analyses, separate propensity score-weighted linear regression models were fitted for each laboratory measurement to evaluate between-group differences in means. Model-based estimated means and corresponding 95% confidence intervals were reported.

Propensity scores were calculated based on age, sex, race/ethnicity, SARS-CoV-2 variant era, referral source, vaccination status at index, comorbidities prior to index, homelessness, employment status, insurance status, income, level of difficulty in covering expenses, last doctor’s visit, and food insecurity to balance on these factors between exposure groups. A complete list of adjustment variables is provided in Table 1. Multiple imputation was used to handle missing data. Propensity score-weighted means, 95% confidence intervals, medians, 10^th and 90^th percentiles were reported for each laboratory measurement. The target populations for estimated model based mean differences in the propensity score model were participants with prior infection in the primary analysis and PASC-positive in secondary analyses. Stabilized weights were applied in all analyses and trimming of top 1% of weights was applied in secondary analyses to address extreme values in the weights.

Table 1a.

Demographics overall and by prior SARS-CoV-2 infection status

Characteristic	Prior Infection (n = 8,746)	No prior infection (n = 1,348)	Overall (n=10,094)
Age at infection [years]
Mean (SD)	46 (15)	52 (15)	47 (15)
Median [IQR]	45 (34, 59)	54 (39, 64)	47 (34, 60)
Missing	5	1	6
Race/Ethnicity
Non-Hispanic White	5080 (58%)	847 (63%)	5927 (59%)
Non-Hispanic Black	1214 (14%)	204 (15%)	1418 (14%)
Non-Hispanic Asian	462 (5%)	88 (7%)	550 (5%)
Hispanic	1523 (17%)	156 (12%)	1679 (17%)
Mixed race/Other	400 (5%)	43 (3%)	443 (4%)
Missing	67	10	77
Sex assigned at birth
Female	6329 (73%)	911 (68%)	7240 (72%)
Male	2392 (27%)	432 (32%)	2824 (28%)
Intersex	4 (0%)	0 (0%)	4 (0%)
Missing	21	5	26
Enrollment group and era
Pre-Omicron	3362 (38%)	317 (24%)	3679 (36%)
Omicron acute	2963 (34%)	548 (41%)	3511 (35%)
Omicron post-acute	2421 (28%)	483 (36%)	2904 (29%)
Referral type
Self-referral/Community outreach/Long COVID	4377 (50%)	960 (71%)	5337 (53%)
Another referral	4364 (50%)	388 (29%)	4752 (47%)
Missing	5	0	5
Hospitalized during acute phase of first infection
Not hospitalized during acute phase	7738 (91%)	n/a	7738 (91%)
Hospitalized during acute phase	723 (9%)	n/a	723 (9%)
Missing	285	n/a	285
Vaccinated at first infection
Unvaccinated	3042 (35%)	182 (14%)	3224 (32%)
Partially vaccinated or date of last dose unknown	516 (6%)	85 (6%)	601 (6%)
Fully vaccinated	5144 (59%)	1068 (80%)	6212 (62%)
Missing	44	13	57
Visit month
6 months	5000 (57%)	952 (71%)	5952 (59%)
9+	3746 (43%)	396 (29%)	4142 (41%)
Marital status at enrollment
Divorced, widowed, separated, or never married	3177 (37%)	531 (40%)	3708 (38%)
Married or living with partner	5376 (63%)	783 (60%)	6159 (62%)
Missing	193	34	227
Homelessness at enrollment
Homeless	197 (2%)	30 (2%)	227 (2%)
Not homeless	8422 (98%)	1294 (98%)	9716 (98%)
Missing	127	24	151
Disability at enrollment
Disabled	319 (4%)	80 (6%)	399 (4%)
Not disabled	8271 (96%)	1233 (94%)	9504 (96%)
Missing	156	35	191
Unemployment at enrollment
Unemployed	245 (3%)	47 (4%)	292 (3%)
Not unemployed	8345 (97%)	1266 (96%)	9611 (97%)
Missing	156	35	191
Medicaid at enrollment
Medicaid	1303 (15%)	176 (13%)	1479 (15%)
Not Medicaid	7196 (85%)	1132 (87%)	8328 (85%)
Missing	247	40	287
Uninsured at enrollment
Uninsured	273 (3%)	38 (3%)	311 (3%)
Not uninsured	8226 (97%)	1270 (97%)	9496 (97%)
Missing	247	40	287
Lost insurance due to pandemic
Lost insurance	274 (3%)	23 (2%)	297 (3%)
Did not lose insurance	8256 (97%)	1279 (98%)	9535 (97%)
Missing	216	46	262
Difficulty covering expenses in the month before enrollment
Not at all difficult to cover expenses	5105 (62%)	933 (73%)	6038 (63%)
Somewhat difficult to cover expenses	2288 (28%)	258 (20%)	2546 (27%)
Very difficult to cover expenses	889 (11%)	92 (7%)	981 (10%)
Missing	464	65	529
Most recent doctor’s visit before index
Within the last 5 years	8373 (99%)	1290 (99%)	9663 (99%)
Greater than 5 years	82 (1%)	15 (1%)	97 (1%)
Missing	291	43	334
Skipped care in 12 months before index
Skipped care	480 (6%)	58 (4%)	538 (5%)
Did not skip care	8035 (94%)	1258 (96%)	9293 (95%)
Missing	231	32	263
2019 Household Income
<$25,000	1319 (16%)	218 (18%)	1537 (17%)
$25,000-$49,999	1279 (16%)	149 (12%)	1428 (15%)
≥$50,000	5403 (68%)	856 (70%)	6259 (68%)
Missing	745	125	870
Food insecurity in 12 months before index
Food insecure	1233 (14%)	156 (12%)	1389 (14%)
Not food insecure	7310 (86%)	1161 (88%)	8471 (86%)
Missing	203	31	234

Sensitivity analyses were performed excluding participants with diabetes (for HgbA1c) and an immunocompromised condition (for platelets). Additional exploratory between group comparisons were participants with prior infection within each PASC cluster vs. those with PASC index of zero.

Results

Participants.

A total of 10,094 participants (8,746 with prior infection; 1,348 uninfected) met study criteria (Figure 1, 72% female [7,240/10,094]; 17% Hispanic/Latino [1,679/10,094]; 14% non-Hispanic Black [1,418/10,094]; 62% fully vaccinated at the index date [6,212/10,037]; median age, 47 years [IQR, 34-60]) (Table 1a). Of participants with prior SARS-CoV-2 infection, 21.5% (1,880/8,746) had PASC index ≥12 and 38.3% (3,351/8,746) as a PASC index of zero (Table 1b). There were no significant differences in demographics and clinical characteristics between participants included and those not included in the analysis cohort (Table S2).

Figure 1:

Consort Diagram

a ) Reinfection window for exclusion is 30 days prior and 7 days after visit

b) Enrollment visit defined as first visit if available, or enrollment date if visit date not available

c) Participants who completed visit without reach end of visit window

Table 1b.

Demographics overall and by PASC Score among individuals with a prior SARS-CoV-2 infection

Characteristic	PASC Score ≥ 12 (n=1880)	PASC Score 0 (n=3351)	Overall (n=5231)
Age at infection [years]
Mean (SD)	48 (14)	47 (16)	47 (15)
Median [IQR]	47 (37, 58)	45 (33, 60)	47 (34, 59)
Missing	1	2	3
Race/Ethnicity
Non-Hispanic White	1209 (65%)	1813 (54%)	3022 (58%)
Non-Hispanic Black	198 (11%)	541 (16%)	739 (14%)
Non-Hispanic Asian	58 (3%)	219 (7%)	277 (5%)
Hispanic	296 (16%)	615 (18%)	911 (18%)
Mixed race/Other	104 (6%)	141 (4%)	245 (5%)
Missing	15	22	37
Sex assigned at birth
Female	1456 (78%)	2242 (67%)	3698 (71%)
Male	419 (22%)	1099 (33%)	1518 (29%)
Intersex	1 (0%)	2 (0%)	3 (0%)
Missing	4	8	12
Enrollment group and era
Pre-Omicron	1203 (64%)	909 (27%)	2112 (40%)
Omicron Acute	317 (17%)	1398 (42%)	1715 (33%)
Omicron Post-Acute	360 (19%)	1044 (31%)	1404 (27%)
Referral type
Self-referral/Community outreach/Long COVID	1180 (63%)	1509 (45%)	2689 (51%)
Another referral	700 (37%)	1840 (55%)	2540 (49%)
Missing	0	2	2
Hospitalized during acute phase of first infection
Not hospitalized during acute phase	1577 (85%)	3011 (94%)	4588 (91%)
Hospitalized during acute phase	272 (15%)	189 (6%)	461 (9%)
Missing	31	151	182
Vaccinated at first infection
Unvaccinated	1058 (56%)	856 (26%)	1914 (37%)
Partially vaccinated or date of last dose unknown	105 (6%)	220 (7%)	325 (6%)
Fully vaccinated	711 (38%)	2254 (68%)	2965 (57%)
Missing	6	21	27
Visit month
6 months	682 (36%)	2218 (66%)	2900 (55%)
9+	1198 (64%)	1133 (34%)	2331 (45%)
Marital status at enrollment
Divorced, widowed, separated, or never married	732 (40%)	1161 (36%)	1893 (37%)
Married or living with partner	1111 (60%)	2103 (64%)	3214 (63%)
Missing	37	87	124
Homelessness at enrollment
Homeless	49 (3%)	69 (2%)	118 (2%)
Not homeless	1813 (97%)	3223 (98%)	5036 (98%)
Missing	18	59	77
Disability at enrollment
Disabled	130 (7%)	80 (2%)	210 (4%)
Not disabled	1729 (93%)	3192 (98%)	4921 (96%)
Missing	21	79	100
Unemployment at enrollment
Unemployed	61 (3%)	85 (3%)	146 (3%)
Not unemployed	1798 (97%)	3187 (97%)	4985 (97%)
Missing	21	79	100
Medicaid at enrollment
Medicaid	341 (18%)	436 (13%)	777 (15%)
Not Medicaid	1505 (82%)	2796 (87%)	4301 (85%)
Missing	34	119	153
Uninsured at enrollment
Uninsured	61 (3%)	105 (3%)	166 (3%)
Not uninsured	1785 (97%)	3127 (97%)	4912 (97%)
Missing	34	119	153
Lost insurance due to pandemic
Lost insurance	136 (7%)	55 (2%)	191 (4%)
Did not lose insurance	1706 (93%)	3206 (98%)	4912 (96%)
Missing	38	90	128
Difficulty covering expenses in the month before enrollment
Not at all difficult to cover expenses	824 (46%)	2211 (70%)	3035 (61%)
Somewhat difficult to cover expenses	603 (34%)	745 (24%)	1348 (27%)
Very difficult to cover expenses	355 (20%)	213 (7%)	568 (11%)
Missing	98	182	280
Most recent doctor's visit before index
Within the last 5 years	1826 (99%)	3174 (99%)	5000 (99%)
Greater than 5 years	16 (1%)	35 (1%)	51 (1%)
Missing	38	142	180
Skipped care in 12 months before index
Skipped care	180 (10%)	110 (3%)	290 (6%)
Did not skip care	1660 (90%)	3149 (97%)	4809 (94%)
Missing	40	92	132
2019 Household Income
<$25,000	299 (17%)	494 (16%)	793 (17%)
$25,000-$49,999	313 (18%)	455 (15%)	768 (16%)
≥$50,000	1130 (65%)	2086 (69%)	3216 (67%)
Missing	138	316	454
Food insecurity in 12 months before index
Food insecure	348 (19%)	409 (13%)	757 (15%)
Not food insecure	1501 (81%)	2854 (87%)	4355 (85%)
Missing	31	88	119

Laboratory measurements associated with prior SARS-CoV-2 infection.

Differences in mean laboratory values between participants with and without prior infection were observed in platelets, HbA1c, and uACR (Table 2, Figure 2) based on propensity score weighted models. Covariate balance was achieved after propensity score weighting (Figure S1). Participants with prior infection had lower mean platelets (265.9 x10⁹cells/L 95% CI = [264.5, 267.4]) compared to participants without known prior infection (275.2 x10⁹cells/L [268.5, 282.0]) and higher mean HbA1c (5.58% [5.56, 5.60] vs. 5.46% [5.40, 5.51]) and uACR (81.9 mg/g [67.5, 96.2]) vs 43.0 mg/g [25.4, 60.6]) compared to participants without known prior infection. Boxplots and empirical cumulative distribution functions of laboratory measurements for HbA1c, platelets and uACR by infection status are provided in Figure S2.

Table 2.

Estimated mean laboratory measurements by group after propensity score weighting

	Prior SARS-CoV-2 Infection	No Prior Infection	PASC score ≥12	PASC score=0
Lab Measurement	Est (95% CI)	Est (95% CI)	Est (95% CI)	Est (95% CI)
Sodium (mmol/L)	139.1 (139.1, 139.2)	139.2 (139.0, 139.4)	139.2 (139.1, 139.3)	139.2 (139.0, 139.4)
Potassium (mmol/L)	4.16 (4.15, 4.16)	4.16 (4.12, 4.19)	4.16 (4.14, 4.18)	4.16 (4.13, 4.19)
Chloride (mmol/L)	103.4 (103.3, 103.5)	103.6 (103.3, 103.9)	103.2 (103.0, 103.3)	103.3 (103.1, 103.6)
Bicarbonate (mmol/L)	25.2 (25.2, 25.3)	25.1 (24.9, 25.4)	25.2 (25.1, 25.3)	25.0 (24.7, 25.2)
Creatinine (serum) (umol/L)	77.5 (76.2, 78.7)	75.4 (72.5, 78.3)	78.4 (75.3, 81.5)	76.7 (73.8, 79.6)
Creatinine (serum) (mg/dL)	0.876 (0.862, 0.890)	0.853 (0.820, 0.886)	0.887 (0.852, 0.922)	0.868 (0.835, 0.900)
Calcium (mmol/L)	2.35 (2.35, 2.36)	2.36 (2.35, 2.37)	2.36 (2.35, 2.36)	2.35 (2.34, 2.36)
Calcium (mg/dL)	9.43 (9.42, 9.44)	9.47 (9.44, 9.51)	9.45 (9.43, 9.47)	9.43 (9.39, 9.46)
ALT (U/L)	22.6 (22.2, 22.9)	22.1 (20.6, 23.6)	23.3 (22.5, 24.2)	22.8 (21.4, 24.1)
AST (U/L)	21.8 (21.5, 22.1)	21.9 (19.9, 23.8)	22.0 (21.3, 22.6)	22.2 (21.1, 23.3)
Total bilirubin (umol/L)	9.56 (9.44, 9.67)	9.45 (9.06, 9.84)	9.09 (8.87, 9.32)	9.13 (8.76, 9.51)
Total bilirubin (mg/dL)	0.559 (0.552, 0.565)	0.553 (0.530, 0.575)	0.532 (0.519, 0.545)	0.534 (0.512, 0.556)
Albumin (g/L)	43.9 (43.8, 44.0)	44.2 (43.8, 44.5)	44.0 (43.9, 44.2)	43.7 (43.4, 44.0)
WBC (x10^9 cells/L)	6.58 (6.53, 6.63)	6.63 (6.42, 6.84)	6.90 (6.80, 7.00)	6.88 (6.62, 7.15)
ALC (x10^9 cells/L)	195.2 (193.3, 197.1)	198.1 (189.3, 207.0)	198.3 (194.9, 201.8)	198.3 (189.7, 206.9)
ANC (x10^9 cells/L)	392.5 (388.7, 396.2)	394.5 (379.1, 410.0)	421.0 (412.1, 429.9)	410.1 (392.8, 427.3)
Hgb (g/L)	136.1 (135.8, 136.4)	137.2 (135.9, 138.5)	136.4 (135.7, 137.1)	136.3 (135.1, 137.6)
Platelets (x10^9 cells/L)	265.9 (264.5, 267.4) (a)	275.2 (268.5, 282.0) (a)	272.7 (269.4, 276.0)	266.3 (259.9, 272.7)
INR	1.02 (1.02, 1.03)	1.02 (1.01, 1.03)	1.02 (1.01, 1.03)	1.02 (0.99, 1.05)
D-dimer (ug/L)	517.6 (497.8, 537.4)	536.8 (481.6, 591.9)	518.6 (476.2, 561.1)	580.0 (471.8, 688.3)
hsCRP (mg/L)	4.08 (3.93, 4.23)	3.65 (3.18, 4.12)	5.01 (4.62, 5.39) (a)	4.23 (3.70, 4.76) (a)
Cystatin-C (mg/L)	0.895 (0.886, 0.904)	0.887 (0.863, 0.911)	0.918 (0.899, 0.937)	0.933 (0.884, 0.981)
HbA1c (%)	5.58 (5.56, 5.60) (a)	5.46 (5.40, 5.51) (a)	5.63 (5.59, 5.67)	5.67 (5.59, 5.74)
HDL cholesterol (mmol/L)	1.50 (1.49, 1.51)	1.49 (1.45, 1.52)	1.47 (1.45, 1.49)	1.47 (1.43, 1.50)
HDL cholesterol (mg/dL)	57.9 (57.5, 58.2)	57.4 (56.0, 58.8)	56.9 (56.1, 57.6)	56.7 (55.4, 57.9)
Non-HDL cholesterol (mmol/L)	3.38 (3.36, 3.40)	3.33 (3.24, 3.41)	3.52 (3.47, 3.57)	3.49 (3.40, 3.59)
Non-HDL cholesterol (mg/dL)	130.7 (129.8, 131.5)	128.6 (125.4, 131.8)	136.2 (134.3, 138.0)	135.0 (131.3, 138.8)
TSH (mcIU/mL)	1.95 (1.91, 1.99)	2.04 (1.85, 2.23)	2.00 (1.90, 2.10)	2.00 (1.86, 2.15)
NT-Pro BNP (pg/mL)	83.7 (76.4, 91.1)	80.8 (64.3, 97.2)	83.1 (67.1, 99.0)	124.6 (51.1, 198.1)
Albumin/creatinine ratio (mg/g)	81.9 (67.5, 96.2) (a)	43.0 (25.4, 60.6) (a)	67.0 (43.5, 90.5)	72.2 (42.7, 101.7)

^(a)Non-overlapping 95% CI between groups.

ALT, alanine transaminase; AST, aspartate aminotransferase; WBC, white blood cell count; ALC, absolute lymphocyte count; ANC, absolute neutrophil count, Hgb, hemoglobin; INR, international normalized ratio; hsCRP, high-sensitivity C-reactive protein; HbA1c, hemoglobin A1c; HDL, high density lipoprotein; TSH, thyroid stimulating hormone; NT-Pro BNP, N-terminal pro b-type natriuretic peptide

Figure 2:

Differences in laboratory measurements between groups after propensity score weighting.

Figure 2A. Prior SARS-CoV-2 infection vs no prior infection

Figure 2B. PASC index ≥12 vs PASC index of zero

Quantiles of propensity score-weighted laboratory values (Table S3) suggest that the right tails of the distributions of HbA1c and uACR were higher in participants with versus without prior infection, while the entirety of the distribution of platelets was shifted lower in participants with versus without prior infection (Table S4).

The difference in mean HbA1c was attenuated after excluding 867 participants with pre-existing diabetes (with prior infection: 5.40% [5.39, 5.42] vs without known infection: 5.37% [5.33, 5.41]). The small difference in mean platelets remained after excluding 640 participants with a pre-existing immunocompromising condition (with prior infection 267.0 x10⁹ cells/L [265.6, 268.5] vs without prior infection: 277.4 x10⁹ cells/L [270.3, 284.6]).

Laboratory measurements associated with a PASC index ≥12 versus zero.

Restricting analysis to participants with prior infection, in propensity score-weighted models balancing on all covariates (Figure S3), we found no clinically meaningful differences (Table 2) in mean laboratory values between those with a PASC index ≥12 versus zero.

Laboratory measurements associated with PASC sub-phenotypes versus PASC index of zero.

In exploratory propensity score-weighted models (Figure S3) restricted to participants with prior infection (Tables S5, S6), we observed: 1) higher mean hsCRP in cluster 1 (smell or taste impairments); 2) lower mean sodium and increase in mean calcium in cluster 2 (high frequency of PEM and fatigue); 3) no differences in cluster 3 (high frequency of brain fog, PEM, and fatigue); and higher hsCRP in cluster 4 (high frequency of fatigue, PEM, dizziness, brain fog, gastrointestinal symptoms, and palpitations), compared to participants with a PASC index of zero. Boxplots and empirical cumulative distribution functions of laboratory measurements by cluster are provided in Figure S4.

Discussion

In a cohort study of over 10,000 participants with and without prior SARS-CoV-2 infection, we found no evidence that any of 25 routine clinical laboratory values provide a reliable biomarker of prior infection, PASC, or specific type of PASC cluster. We did identify small differences in some laboratory tests between those with and without prior infection: specifically, we found that prior infection was associated with modest increases in HbA1c and uACR, and small decreases in platelets. Within PASC symptom clusters, hsCRP was slightly higher in cluster 1 (impairments in smell and taste) and cluster 4 (high frequency of fatigue, PEM, dizziness, brain fog, gastrointestinal symptoms, and palpitations), while sodium was lower and calcium higher in cluster 2 (high frequency of PEM and fatigue), compared to participants with a PASC index of zero. These results may have been due to chance given multiple comparisons, and were largely not clinically meaningful. Thus, while clinicians should rule out treatable causes of PASC symptoms with appropriate diagnostic testing, routine laboratory tests are not useful biomarkers for PASC. Earlier, smaller studies have found similar results (27). While these small differences are not useful for PASC diagnosis, they may suggest potential pathways in the pathogenesis of PASC and PASC clusters.

First, an association between severity of acute COVID-19 with diabetes or glucose intolerance has been recognized since early in the COVID-19 pandemic. SARS-CoV-2 itself may contribute to the development of diabetes and previously undiagnosed diabetes or receipt of high-doses of corticosteroids during the hospital stay may play a role in the severity of initial disease (29). The association between diabetes and PASC has been less well studied, and often limited to electronic health record data of patients with variably-defined PASC and shorter follow-up duration after initial infection. In a cohort of over 5 million patients with and without prior SARS-CoV-2 infection, prior infection was associated with increased risk of subsequent diabetes (30). A study of nearly 10 million veterans found that veterans with a SARS-CoV-2 diagnosis had an increased risk of incident diabetes and antihyperglycemic use in the 12 months after diagnosis; however, 15% of the cohort were missing HbA1c and there may have been underdiagnosis of antecedent diabetes, particularly in the control groups (31). Our results are consistent with both of these large cohort studies, with the advantage of systematic HbA1c collection and comprehensive, standardized symptom collection. Of note, we did not detect a difference in HbA1c when those with pre-existing diabetes were excluded in the sensitivity analysis. Furthermore, we compared HbA1c by PASC index severity. While in theory, SARS-CoV-2 infection may increase risk of diabetes, diabetes may also worsen PASC symptomatology, including microvascular complications (32). Despite our large cohort size, we did not detect a difference in HbA1c between those with PASC index of ≥12 and those with PASC index of zero, suggesting that SARS-CoV-2 may contribute to glucose dysregulation independent of symptoms.

A second finding distinguishing participants with and without prior SARS-CoV-2 infection was the shift towards lower platelets, although our detectable platelet differences were likely of minimal clinical relevance, and most participants were well within the clinically normal range. In contrast, among participants with PASC index ≥12, both platelets and hsCRP tended to be higher compared to those with PASC index of zero, suggesting an ongoing inflammatory state, consistent with prior literature (33-36). Platelet and clotting abnormalities are more complex than simply the total number of platelets, and abnormalities have been recognized widely during acute COVID-19, including in vitro studies demonstrating internalization of SARS-CoV-2 virion by platelets and rapid cell death (37). Some evidence supports a role of clotting and platelet abnormalities in PASC (38), though other studies have not found convincing evidence of microclots (39, 40). Low platelets seen in association with prior infection may be multifactorial, reflecting platelet destruction, marrow suppression, or consumption, as seen with other viral infections (e.g. HIV, Epstein-Barr virus). Indeed, one study found downregulation of platelet genes among patients with Long COVID (41). Other studies have demonstrated markers of platelet activation in PASC (42). Among a subset of 80 symptomatic participants in a Long COVID/PASC registry, microthrombi and platelet abnormalities (hyperactivity) were seen in nearly all participants (38, 43). Platelet abnormalities have also been described among patients with myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) or postural orthostatic tachycardia syndrome (POTS) (44, 45). Clinical quantitation of platelet numbers is not a reliable biomarker for PASC and more specific markers of platelet biology will likely be needed to detect platelet dysfunction related to PASC.

The other difference among participants with and without prior SARS-CoV-2 was a higher uACR, without differences in cystatin C or creatinine (46). Smaller case-control studies have found increased uACR either acutely or several months following SARS-CoV-2, compared to those without infection (47, 48). Other larger observational studies have shown declines in renal function in the year following SARS-CoV-2 infection, with a slightly greater proportion of individuals with moderately or severely increased uACR from 3 to 12 months, though no statistical comparisons were provided (46, 49, 50). These findings may reflect greater comorbidity among those with prior infection or consequences of severe initial infection; other studies have suggested a direct effect of SARS-CoV-2 on renal tubules and parenchyma, resulting in renal dysfunction (51, 52). Beyond the direct association with renal function, elevated uACR (even in the microalbuminuria range) has been associated with increased cardiovascular disease risk (53, 54). In HIV, albuminuria is associated with increased mortality, cardiovascular disease, and heart failure (55) through endothelial dysfunction (56, 57), a known contributor to PASC-related cardiovascular risk (58) that will be assessed in future RECOVER analyses.

The association with hsCRP and cluster 1 (smell/taste disturbances) is consistent with a recent study that looked at the ultrastructural changes of olfactory bulbs and tracts in SARS-CoV-2 infection (59). The authors suggested that anosmia is likely due to inflammation from the virus, rather than direct viral invasion. The association with hsCRP and smell/taste disturbance conflicts with a recent systematic review of 11 different studies in which patients with chemosensory disturbances tended to have lower overall inflammation, including hs-CRP, compared to those without chemosensory disturbances (60). This discrepancy may reflect the newer variants captured in our cohort in comparison to many studies with earlier variants where chemosensory disturbances were often associated with lower disease severity (61).

Our study has many strengths as the first large study that prospectively assessed a broad battery of clinical laboratory biomarkers with standardized systematic evaluation for symptoms, reducing ascertainment bias. We had large and robust control groups, and our participants are diverse in terms of demographics, geographic distribution, variant of infection and vaccination status. We use a rigorously derived, concrete and reproducible research definition of PASC, enabling this cohort to be well characterized. However, our study also has important limitations. Laboratory studies were run at local CLIA-certified labs whose assays may differ slightly from each other. Laboratory measures were not always drawn fasting. We did not have pre-infection laboratory results on most participants, impeding ability to study pre/post infection changes and limiting the distinction between abnormalities predisposing to infection and those resulting from infection. We considered each measure independently and did not evaluate multivariate laboratory profiles. The analysis combined laboratory values and PASC status assessed between a 6-month visit (180 +/− 45 days) and up to 3 years after index date. Finally, other clinical laboratory studies not captured here (e.g., cortisol, fibrinogen) warrant further study.

Conclusions

While clinicians should obtain routine clinical tests to rule out other treatable causes of PASC symptoms, we found no evidence that 25 routine clinical laboratory values offer clinical utility as biomarkers of PASC, and are therefore not useful as part of a definition. We did find slightly lower platelets, higher HbA1c, and higher uACR among participants with prior SARS-CoV-2 infection compared to those without. Whether these differences represent consequences of or risk factors for initial acquisition or severity of SARS-CoV-2 infection cannot be determined without pre-infection evaluations. Furthermore, differences were quite small, likely of minimal clinical relevance, and may have been due to chance. Laboratory differences among those within PASC subphenotypes suggest ongoing inflammation (hsCRP) as a potential mechanism underlying PASC. Future work within the RECOVER cohort will evaluate whether clusters of biomarkers can help refine a research-based or clinically relevant PASC diagnosis.

In summary, our findings suggest that even highly symptomatic PASC may have no clinically observable objective findings on routine laboratory testing. Understanding of the basic biologic underpinnings of persistent symptoms after SARS-CoV-2 infection will likely require a rigorous focus on investigations beyond routine clinical laboratory studies (e.g., transcriptomics, proteomics, metabolomics) to identify novel biomarkers.