Harmonization of Four Biomarkers Across Nine Nationally Representative Studies of Older Persons

Abstract

Introduction

A growing number of international population surveys have included measurement of biomarkers, but differ in the type of specimens collected, sample processing procedures, shipment protocols, and laboratory assay platforms. The purpose of this study is to harmonize biomarker data from nine nationally-representative studies of people 50 years of age and over by adjusting for assay platforms and type of specimens for total cholesterol (total-C), high-density lipoprotein cholesterol (HDL-C), glycosylated hemoglobin (HbA1c), and C-reactive protein (CRP).

Methods

Sets of 24 identical serum, plasma, whole blood, and dried blood spot harmonization samples with known analyte levels were generated at a reference laboratory, shipped at −80°C to the respective study laboratories, and subsequently assayed following the study laboratory’s protocol. Both original and harmonized study data were used to calculate mean values and at-risk prevalence.

Results

The correlation coefficients between the biomarker values of the harmonization samples obtained by the study laboratories and the reference laboratory were 0.99 or above for all analytes and laboratories, indicating the high quality of assays at all laboratories. However, using the harmonized data from each study, there were significant differences in the mean values and country ranking of the prevalence of at-risk levels of these four biomarkers.

Conclusions

While the biomarker data from the different study laboratories were highly correlated, indicating very high correlation of rank order of specimens, absolute values did vary significantly. This can have a major impact on assessment of international differences in estimates of risks for chronic morbidity and mortality.

Introduction

A growing number of international population studies have included measurement of biomarkers to better understand age-related physiological changes, diseases that have onset linked to older age, and the aging process itself.¹ Many of the aging surveys are harmonized with the US Health and Retirement Survey (HRS) in multiple domains, including education, income, wealth, health, transfers, and expectations. They have also made biological data publicly available^2–9 and offered opportunities to conduct rigorous, international comparisons of health outcomes to gain insight into the determinants and dynamics of health status.¹⁰ In particular, this type of cross-country analysis allows examination of some institutional or policy factors, such as the organization of health insurance and the provision of health care by welfare states or through workplace arrangements, that often vary more systematically and perhaps exogenously across countries than within countries.¹¹

However, these population surveys differ significantly in the type of specimens collected (e.g., venous blood versus dried blood spots), sample processing procedures, shipment protocols, and laboratory assay platforms. The variations in these practices could result in different levels and distributions of the biomarkers being tested, making direct comparisons of biomarker data potentially inaccurate. This could be particularly problematic when diseases or pathophysiological states are defined by clinically accepted cut-off points, such as using glycosylated hemoglobin (HbA1c) of 6.5% and greater to diagnose diabetes mellitus.¹²

In laboratory medicine, “standardization” is used when laboratory results for a measurement are equivalent through a high-order primary reference material and/or a reference measurement procedure, while “harmonization” is generally used when laboratory results are equivalent, but neither a high-order primary reference material nor a reference measurement procedure is available.¹³ The goal of harmonization to ensure that the assay information provided is comparable, irrespective of the measurement procedure used and where and/or when a measurement is made. This process often involves development of commutable secondary reference materials.¹⁴

In this study, we have harmonized assays of four blood-based biomarkers for selected international surveys in the HRS family of studies: total cholesterol (total-C), high-density lipoprotein cholesterol (HDL-C), glycosylated hemoglobin (HbA1c), and C-reactive protein (CRP) for nine population-representative international studies. This was accomplished by sending a replicate set of harmonization samples to each study laboratory for measurement of these four biomarkers. In this analysis, we have used both the original and harmonized biomarker data to describe cross-national differences and compare cross-national differences in relationships between these biomarkers with and without harmonization.

Materials and Methods

Studies

We included the nine studies listed in Table 1, which are all members of the HRS family of studies. The sister studies in our analysis include the English Longitudinal Study of Ageing (ELSA), the Irish Longitudinal Study on Ageing (TILDA), the China Health and Retirement Longitudinal Study (CHARLS), the Northern Ireland Cohort for the Longitudinal Study of Ageing (NICOLA), the Indonesia Family Life Survey (IFLS), the Brazilian Longitudinal Study of Aging (ELSI), and the Longitudinal Aging Study in India (LASI) and its substudy, the Diagnostic Assessment of Dementia (LASI-DAD) (https://hrs.isr.umich.edu/about/international-sister-studies).¹⁵ The HRS family surveys were designed and are coordinated with the explicit goal of facilitating cross-country comparisons. Like HRS, most or all of these studies have (1) biennial interviews with respondents and their spouses; (2) a multidisciplinary questionnaire design that elicits a wealth of information about health, demographics, and other topics; and (3) regular refreshment samples to keep the sample representative of the older population.

Table 1:

Studies included in harmonization, the analyzers used, sample types collected by each study (with comparable sample types used for the calibration samples) & biomarkers assayed

Study Name	Country	Manufacturer		Sample Type				Biomarkers
		Chemistry Analyzer	HbA1c Analyzer	Whole blood	Plasma	Serum	Dried blood spots	HbA1c*	TC*	HDL-C*	CRP*
CHARLS	China	Abbott Architect	N/A		X				X	X	X
ELSA	England	Roche Cobas 8000	TOSOH G8	X		X		X	X	X	X
ELSI-Brazil	Brazil	Siemens Advai 2400	Bio-Rad Variant Turbo	X		X		X	X	X
HRS	United States	Roche Cobas 6000	N/A			X			X	X	X
IFLS	Indonesia	Bio-Rad iMark	Bio-Rad D-10				X	X			X
LASI	India	Tecan Sunrise	Roche Cobas 400				X	X			X
LASI-DAD	India	Abbott Architect	Bio-Rad D-10	X		X		X	X	X	X
NICOLA	Northern Ireland	Werfen ILab 600 (Ireland lab)	Werfen ILab 600 (Ireland lab)	X		X		X	X	X	X
		Abbott ARCHITECT c8000 (Germany Lab)
TILDA	Ireland	Roche Cobas c701	ADAMS A1c HA-8180V	X	X			X	X	X	X

^*HbA1c: Hemoglobin A1c; TC: Total Cholesterol; HDL-C: High-density lipoprotein cholesterol; CRP: C-reactive protein

“x” indicates the specific Sample Type and Biomarkers collected and assayed by each study.

HRS=Health and Retirement Study, ELSA=English Longitudinal Study of Ageing, TILDA=Irish Longitudinal Study on Ageing, NICOLA=Northern Ireland Cohort for the Longitudinal Study of Ageing, CHARLS=China Health and Retirement Longitudinal Study, LASI=Longitudinal Aging Study in India, LASI-DAD=Longitudinal Aging Study in India – Diagnostic Assessment of Dementia Study, IFLS=Indonesia Family Life Survey, ELSI=Brazilian Longitudinal Study of Aging.

All of these studies have collected biomarker data using a variety of sample types in the assays, whole blood, serum, plasma, and dried blood spots (DBS), as well as a variety of analyzers to conduct the assays (Table 1). Four of the studies performed all four of the assays that we harmonized here [ELSA (England), LASI-DAD (India), NICOLA (Northern Ireland), and TILDA (Ireland)]; HRS (USA) did not measure HbA1c; ELSI (Brazil) did not measure CRP; and the two studies using DBS – LASI (India) and IFLS (Indonesia) – did not measure either total-C or HDL-C. Also, we were unable to calibrate HbA1c data for CHARLS (China) because of an inability to obtain administrative approval from the Chinese government to ship whole blood specimens to China. Finally, we were unable to apply our results for HbA1c and CRP to study samples for TILDA (Ireland) because those data are not yet released.

The characteristics of the data that we compared across countries are shown in Table 2. All data were collected within a few years of each other, and we limited the study populations included in analyses to participants over age 50, or to the lower end of the age of collection if it was greater than 50, in order to compare study populations of fairly similar age. However, study populations did vary somewhat in age, e.g., the mean is 61.3 years for IFLS (Indonesia) and 69.5 years for LASI-DAD (India), which is a survey focused on dementia.

Table 2:

Characteristics of Studies harmonized for this analysis

Study	Age Eligible (years)	Sample Size	Percent	Year of Collection	Mean Age (years)	% Female
CHARLS	50+	8,510	10.25	2015	63.1	51.3
ELSA	50+	2,710	3.26	2016–2017	67.7	52.5
ELSI	50+	2,361	2.84	2015–2016	62.6	53.9
HRS	56+	9,189	11.07	2016	68.7	54.2
IFLS	50+	2,600	3.13	2014–2015	61.3	53.0
LASI	50+	46,517	56.02	2017–2019	62.0	49.8
LASI-DAD	60+	2,892	3.48	2017–2019	69.5	50.8
NICOLA	50+	3,424	4.12	2017	66.1	49.7
TILDA	53+	4,829	5.82	2014–2015	66.4	53.4
TOTAL		83,032	100

HRS=Health and Retirement Study, ELSA=English Longitudinal Study of Ageing, TILDA=Irish Longitudinal Study on Ageing, NICOLA=Northern Ireland Cohort for the Longitudinal Study of Ageing, CHARLS=China Health and Retirement Longitudinal Study, LASI=Longitudinal Aging Study in India, LASI-DAD=Longitudinal Aging Study in India – Diagnostic Assessment of Dementia Study, IFLS=Indonesia Family Life Survey, ELSI=Brazilian Longitudinal Study of Aging.

Creation, Transportation, and Assay of Harmonization Samples

Multiple replicate sets of 24 serum, plasma, whole blood, and DBS samples were created by the Biomarker Laboratory at the Department of Laboratory Medicine and Pathology, University of Washington (UW) for use in each of the four assays being harmonized, i.e., total-C, HDL-C, HbA1c, and CRP.

Separately for total-C, HDL-C and CRP, to produce serum or plasma samples with different analyte concentrations, a harmonization sample with the highest desired analyte concentration was created by adding an aliquot of an analyte concentrate to a volume of as-received human serum or plasma. Harmonization samples with successively lower desired analyte concentrations were then created by adding a volume of the highest analyte concentration sample to an increasing volume of the as-received human serum or plasma. In the event that the as-received serum or plasma had an analyte concentration greater than lower desired analyte concentrations, successively lower analyte concentration harmonization samples were created by adding increasing volumes of an analyte-negative human albumin solution to a volume of the as-received serum or plasma. To produce whole blood harmonization samples for the HbA1c assay, venous blood samples were pooled by measured HbA1c value to create a group of samples spanning from the highest to the lowest desired HbA1c value.

The two studies that collected DBS (India LASI and Indonesia) only measured CRP and HbA1c levels on study specimens. Aliquots (0.75 uL) from each of the whole blood HbA1c harmonization samples (above) were spotted onto Whatman 903 filter paper strips to create DBS harmonization samples spanning the desired range of HbA1c values. The UW DBS HbA1c assay was used to establish the HbA1c value of each DBS harmonization sample. For the CRP assay, a fixed volume of each of the plasma CRP harmonization samples (above) was mixed with a fixed volume of washed human erythrocytes suspended in a human albumin solution; 0.75 uL aliquots of each mixture were then spotted onto Whatman 903 filter paper strips to create DBS harmonization samples spanning the desired range of CRP concentrations. The UW DBS CRP assay was used to establish the CRP value of each DBS harmonization sample. Since the liquid harmonization samples (above) were used to create the DBS harmonization samples, each DBS specimen is directly associated with the “conventional” whole blood HbA1c or plasma CRP specimen. Study-specific linear regressions of the assigned whole blood HbA1c or plasma CRP concentration on the harmonization DBS HbA1c or DBS CRP were used to convert the raw IFLS and LASI DBS results into harmonized whole blood HbA1c units or plasma CRP units. This long-standing practice permits the use of clinical metrics in the evaluation of data obtained from analyses of DBS samples.¹⁶

The ranges of analyte concentrations were 0 to 400 mg/dL for total-C, 0 to 130 mg/dL for HDL-C, 0 to 30 mg/L for CRP, 3.4% to 14.0% for HbA1c, 0 to 24 mg/L for DBS CRP, and 4.2% to 12.2% for DBS HbA1c. UW measured total-C, HDL-C, and CRP levels on a Beckman Olympus AU680 Chemistry Analyzer (Beckman Coulter, Brea, CA), DBS CRP by BioCheck ELISA (Biocheck Inc., South San Francisco, CA), and HbA1c levels on a Bio-Rad Variant II HPLC (Bio-Rad Laboratories, Hercules, California), which served as the “gold” standard for our harmonization efforts.

To each study laboratory, we sent the type of harmonization sample that matched the type of sample collected by the national study; all samples were shipped at −80°C using World Courier, a commercial cold chain shipping company. A temperature monitor was included with each of the shipments, which recorded once every two hours the temperature inside the shipping box. The subsequent analysis of the data from the monitors showed that temperature was maintained between −70°C to −80°C for all shipments. After arrival at the laboratories, the harmonization samples were stored at −80°C until the samples were assayed.

All study laboratories followed the same workflow protocol by testing the harmonization samples over a four-day period. Samples with identification numbers 1 through 12 were thawed on Day 1 and assayed immediately. These 12 samples were then stored overnight at 4°C and assayed again on Day 2. Samples with identification numbers 13 through 24 were assayed in the same way on Days 3 and 4.

Methods of Analysis

We applied a “classic calibration” approach to harmonize the data across studies. The “classic calibration” is a standard method used in laboratory medicine to convert results from one method to another,¹⁷ for example when assays are discontinued and must be replaced. Here we harmonized the biomarker values of each study’s samples to UW values. UW was treated as the reference laboratory in that the biomarker data obtained from UW’s assays of the harmonization samples were compared with the biomarker data obtained from the replicate harmonization samples assayed in duplicate at each of the study laboratories. We assessed the similarity of the relationships between the UW values and the values from the study laboratories by comparing the correlations between UW and each individual study laboratory. Using log-transformed data from the harmonization samples, we developed a harmonizing equation for each study laboratory in the following format.

$$\text{Study\hspace{0.17em}Laboratory\hspace{0.17em}Harmonization\hspace{0.17em}Sample}={\widehat{\beta }}_0+{\widehat{\beta }}_1\text{UW}$$

To permit comparisons between the national studies, we then applied the coefficients from that equation (shown in Table 3) for each study laboratory to the log-transformed original raw study data to estimate a harmonized version of the study data in the following equation.

Table 3.

Harmonization equations applied to log-transformed regression equations of study data

	Total Cholesterol		HDL Cholesterol		Hemoglobin A1c		C-reactive Protein
Study	Slope	Intercept	Slope	Intercept	Slope	Intercept	Slope	Intercept
CHARLS	1.07	−0.187	1.11	−0.37	NA	NA	0.941	0.084
ELSA	1.00	−0.038	1.07	−0.217	0.98	0.046	0.907	0.07
ELSI	0.995	−0.003	0.942	0.077	1.04	0.003	NA	NA
HRS	1.02	−0.038	0.997	0.021	NA	NA	0.922	0.106
IFLS	NA	NA	NA	NA	0.958	0	0.762	0.431
LASI	NA	NA	NA	NA	1.01	0.044	0.826	0.232
LASI-DAD	1.08	−0.197	1.05	−0.191	1.01	0.034	0.834	0.24
NICOLA	0.999	0.011	1.02	−0.145	1.00	0.033	0.90	0.136
TILDA	1.04	−0.137	1.01	−0.018	0.958	0.049	0.90	0.038

HRS=Health and Retirement Study, ELSA=English Longitudinal Study of Ageing, TILDA=Irish Longitudinal Study on Ageing, NICOLA=Northern Ireland Cohort for the Longitudinal Study of Ageing, CHARLS=China Health and Retirement Longitudinal Study, LASI=Longitudinal Aging Study in India, LASI-DAD=Longitudinal Aging Study in India – Diagnostic Assessment of Dementia Study, IFLS=Indonesia Family Life Survey, ELSI=Brazilian Longitudinal Study of Aging.

$$\text{Harmonized\hspace{0.17em}Study Value}=(\text{Study\hspace{0.17em}Raw\hspace{0.17em}Value}-{\widehat{\beta }}_0)/{\widehat{\beta }}_1$$

Data were thus all harmonized to UW values. This does not mean that the “real” value was the UW harmonized value, but that now all studies were on the same assay scale. These harmonized values could be compared as if all study samples had been assayed at UW and hence we could see the effect of this harmonization on the differences between studies.

We examined the effect of harmonization on the values of each biomarker by comparing the estimates of the mean values of the original raw study data and the harmonized study data from each country. We also estimated the prevalence of high-risk values of the analytes as these were often how data on these biomarkers were used in analyses and because harmonization could affect the break points defining high risk differently from the effect on the mean. High-risk categories were defined as total-C ≥ 240 mg/dL, HDL-C < 40 mg/dL, HbA1c ≥ 6.5%, and CRP ≥ 3 mg/L. The study data were weighted using study weights.

Results

There were very high correlations between the biomarker values measured in the harmonization samples by all study laboratories and UW; all correlation coefficients between the values measured by an individual study laboratory and UW were 0.99 or 1.00. This means that all the laboratories essentially produced the same rank order of the harmonization samples, which indicates that each of the laboratories produced high quality data for their studies. The equations reflecting the association by analyte between the values of the harmonization samples obtained by UW and by each study laboratory are shown in Table 3.

These equations were applied to the raw study data to produce new individual values for each study sample so that each study then had data that were harmonized to UW values. The two sets of data, raw and harmonized, were compared to determine how the harmonized study data differed from the raw data.

Total Cholesterol (total-C).

After applying the harmonization regression equations, the results indicated that the raw data produced by many of the laboratories differed from the harmonized data, leading to changes in the ranking of country values (Figure 1a and Table S1). Harmonizing the data resulted in increased values of mean total-C for all studies except NICOLA (Northern Ireland) and HRS (USA). The changes were quite dramatic for ELSA (England, 18.4 mg/dL) and TILDA (Ireland, 19.9 mg/dL); and somewhat less so for CHARLS (China, 11.4 mg/dL), ELSI (Brazil, 6.5 mg/dL), and LASI-DAD (India, 6.1 mg/dL) (Table S1). In the raw data, NICOLA (Northern Ireland) had the highest total-C and LASI-DAD (India) the lowest. In the harmonized data, HRS (USA) had the lowest total-C and ELSA (England) had the highest. In the harmonized data, the percent high-risk increased for all countries except HRS (USA) and NICOLA (Northern Ireland, Table 4). The increase was especially notable for ELSA (England, 16.4%) and TILDA (Ireland, 10.0%).

Figure 1.

Mean values for raw and harmonized data

Table 4:

Percent with high risk levels of each biomarker in raw and harmonized study data

	High Risk Total Cholesterol (≥ 240 mg/dL) (N=33,320)				High Risk HDL Cholesterol (< 40 mg/dL) (N=33,587)				High Risk HbA1c (≥ 6.5%) (N=59,713)				High Risk CRP (≥ 3 mg/L) (N=75,285)
	Raw		Harmonized		Raw		Harmonized		Raw		Harmonized		Raw		Harmonized
	%	95% CI	%	95% CI	%	95% CI	%	95% CI	%	95% CI	%	95% CI	%	95% CI	%	95% CI
CHARLS	6.94	6.03, 7.86	11.16	9.99, 12.32	16.41	14.90, 17.92	0.14	0.05, 0.23	NA	NA	NA	NA	22.52	21.04, 24.01	18.83	17.39, 20.27
ELSA	18.00	16.29, 19.72	34.36	32.18, 36.53	10.91	9.36, 12.45	2.65	1.78, 3.51	10.54	9.00, 12.08	6.62	5.44, 7.80	27.53	25.43, 29.62	25.53	23.49, 27.57
ELSI	11.42	9.34, 13.51	15.75	13.38, 18.12	32.13	28.74, 35.52	29.23	25.97, 32.69	16.58	14.26, 18.90	10.31	8.43, 12.18	NA	NA	NA	NA
HRS	12.11	11.23, 12.98	9.65	8.86, 10.44	15.59	14.63, 16.55	19.75	18.70, 20.81	NA	NA	NA	NA	38.60	37.34, 39.86	33.22	32.00, 34.44
IFLS	NA	NA	NA	NA	NA	NA	NA	NA	3.75	2.93, 4.56	5.39	4.41, 6.37	30.25	28.19, 32.30	11.89	10.46, 13.32
LASI	NA	NA	NA	NA	NA	NA	NA	NA	40.92	40.36, 41.48	15.85	15.44, 16.26	15.92	15.51, 16.34	10.07	9.73, 10.41
LASI-DAD	7.91	6.78, 9.04	8.62	7.44, 9.81	36.71	34.56, 38.86	7.79	6.59, 8.99	19.46	17.75, 21.17	11.92	10.54, 13.30	34.91	32.77, 37.05	24.93	22.97, 26.88
NICOLA	29.25	27.65, 30.85	26.11	24.56, 27.65	4.99	4.25, 5.73	0.41	0.19, 0.62	10.82	9.69, 11.94	6.08	5.21, 6.95	30.62	29.06, 32.17	24.37	22.92, 25.81
TILDA	7.5	7.62, 9.36	17.5	17.89, 20.29	15.9	15.81, 18.20	15.5	15.81, 18.20	NA	NA	NA	NA	NA	NA	NA	NA

HRS=Health and Retirement Study, ELSA=English Longitudinal Study of Ageing, TILDA=Irish Longitudinal Study on Ageing, NICOLA=Northern Ireland Cohort for the Longitudinal Study of Ageing, CHARLS=China Health and Retirement Longitudinal Study, LASI=Longitudinal Aging Study in India, LASI-DAD=Longitudinal Aging Study in India – Diagnostic Assessment of Dementia Study, IFLS=Indonesia Family Life Survey, ELSI=Brazilian Longitudinal Study of Aging.

HDL Cholesterol (HDL-C).

As was noted for total-C, the harmonized values of HDL-C were also higher than the raw values, except in HRS (USA). TILDA (Ireland) changed very little (0.03 mg/dL) but CHARLS (China, 23.2 mg/dL) and NICOLA (Northern Ireland, 17.3 mg/dL) increased substantially (Figure 1b and Table S1). Comparing raw values of HDL-C, ELSA (England) and NICOLA (Northern Ireland) had the highest values and LASI-DAD (India) the lowest, but after harmonization, NICOLA (Northern Ireland) had the highest values and ELSI (Brazil) the lowest.

The harmonized levels of high-risk HDL-C were reduced from the raw values for all countries except HRS (USA, Table 4). The levels for CHARLS (China), ELSA (England) and NICOLA (Northern Ireland) were very low. There was no significant change for ELSI (Brazil) or TILDA (Ireland). The largest reduction was from 16.41% to 0.14% for CHARLS (China), from 4.99% to 0.41% for NICOLA (Northern Ireland), and from 36.71% to 7.79% for LASI-DAD (India). We should note that we used 40 mg/dL as the cut-off for both men and women which could underestimate the true prevalence of at-risk HDL-C for women, since several clinical practice guidelines have suggested using 50 mg/dL to define increased risk for women.^18,19

HbA1c.

There was some reduction in the mean values of HbA1c from the raw to harmonized data for all countries except IFLS (Indonesia, Figure 1c and Table S1). The harmonization did not change the country with the highest mean value (LASI - India) or the lowest (IFLS - Indonesia). The change in the percent having a high-risk level of HbA1c was exceptionally large for LASI (India), a decrease of 25.07 percentage points (Table 4). Even with the decrease, the LASI (India) study had the highest level of elevated HbA1c, as it did before harmonization; the study with the lowest level of risk, IFLS (Indonesia), also remained the same.

C-Reactive Protein (CRP).

For CRP, the means of the harmonized data from IFLS (Indonesia), LASI (India) and NICOLA (Northern Ireland) were outside the 95% confidence interval of the means of the raw study data and they all decreased (Figure 1d and Table S1). The mean value of CRP for IFLS (Indonesia) when harmonized was reduced so that it became the country with the lowest value; the mean value for LASI-DAD (India) was highest both before and after harmonization.

When the data were harmonized, the level of high-risk level CRP was generally reduced, except ELSA (England) for which there was no significant change (Table 4). There was a very large reduction in the value for IFLS (Indonesia, from 30.25% to 11.89%). HRS was the study with the highest percentage of at-risk level CRP, and LASI (India) the study with the lowest, both before and after harmonization.

Discussion

Our data indicated that each laboratory used by the population-based studies included in this harmonization project generated total and HDL cholesterol, HbA1c, and CRP data from the harmonization samples that were highly correlated with the values from the reference laboratory at the University of Washington. This indicated that all study laboratories had performed these assays well and therefore that the relative ranking of the studies by the mean analyte value of study samples was accurate. However, differences between the performance of the assays conducted at each laboratory, as shown by the differences between the harmonization equations (Table 3), were sufficient to cause the estimated mean and prevalence of high-risk values in a national study to change significantly after harmonization. For example, after harmonization the prevalence of high-risk total cholesterol in the ELSA (England) study increased by 16.36%, leading to a different relative ranking across the countries (Table 4). These changes after data harmonization have important public health and policy implications, because prevalence is a measure of disease burden and is often used for health care planning, resource allocation, and evaluation of intervention effectiveness at local and country levels.²⁰

The ability to perform accurate measurements that are comparable over time and across laboratories is essential for ensuring appropriate clinical and public health practice. It has been recognized that, even when the laboratories use assays consisting of calibrators, reagents, and instruments from the same manufacturer, the assay results may not always be accurate and interchangeable. An independent external quality assessment, such as standardization of cholesterol measurement by the National Cholesterol Education Program (NCEP) and standardization of HbA1c measurement by the National Glycohemoglobin Standardization Program (NGSP), plays a key role in ensuring optimal patient care and public health.²¹ However, even with the dedicated effort by the NCEP and NGSP, there remain significant variations in assay results because NCEP performance criteria allow 8.9% total error for total-C and 13% total error for HDL-C, and NGSP performance criteria allow 5% total error for HbA1c.^22,23

The higher acceptable imprecision of HDL-C values than of total-C values is consistent with our findings, which showed more dramatic changes in mean HDL-C levels and in the prevalence of samples in the at-risk category after harmonization. A previous study that compared assay results from 63 chemistry analyzers from 6 different manufacturers also found up to 8% difference in HDL-C levels and frequent concentration-related biases.²⁴

Factors that can affect the comparability of results across laboratories include differences between the standards of performance used by national and international agencies to accredit laboratories,²⁵ differences between accreditation program requirements,²⁶ and the lack of a universal requirement that a clinical laboratory participate in a quality assurance program.^27,28 Recent activities to reduce the impacts of these issues including regionally adopting common laboratory performance standards,²⁹ addressing differences between standards and quality assurance programs,²⁶ and requiring clinical laboratories to be accredited³⁰ are expected to improve the compatibility of test results across laboratories. The findings from this biomarker harmonization study highlight the necessity for assay value harmonization not only across studies but also when biomarker data are compared between different waves of the same longitudinal study, which may have made changes in study laboratory and assay methodology over time. This harmonization effort should be implemented in real-time, when the study specimens are being measured. It would also be helpful to develop a more coordinated mechanism at a reference laboratory to assist individual studies to accomplish this goal by providing appropriate harmonization samples and reference values. Interest in improving standardization and harmonization across laboratories has led to promulgation of guidance on how to establish such a program.³¹

This study has several important strengths. Different types of harmonization samples were created to match the exact type of specimens collected by the studies, including serum, plasma, whole blood, and DBS, and were stored under −80°C during storage and shipment to maintain sample stability. This allowed more accurate harmonization between study biomarker data and our reference laboratory. In addition, the harmonization samples generated covered the usual range of analyte variations to minimize the need for data extrapolation. Limitations of this study should also be noted. First, we have focused on harmonization of assay performance across study laboratories, but were not able to fully address the impact of specimen collection, processing, and shipment has on biomarker results. For example, values from DBS-based assays might have been influenced by spot size, drying time, placement of desiccant and temperature exposure during sample shipment.³² High humidity could be another factor that may affect DBS-based assay results. Both studies in India (LASI) and Indonesia (IFLS) dried DBS cards for at least 4 hours or overnight before storage in freezers. Moreover, previous research showed that the effect of high humidity on DBS samples varies by analyte, with CRP and HbA1c results being relatively insensitive to humidity levels.³³ Delayed separation of plasma from whole blood could also affect biomarker results, although the change in cholesterol levels is typically no more than 5%.³⁴ Second, we did not examine some factors that might potentially affect at-risk prevalence, such as medication use and prevalence of various comorbid conditions. Lastly, the biomarker data we have harmonized were measured on chemistry analyzers, HbA1c analyzers, or using commercially available ELISA kits. It is possible that other assay platforms are more prone to error and technical variations.

Conclusions

Our study has shown that, while the biomarker data from different national studies within the HRS family were highly correlated, absolute values did vary significantly. This can have a major impact on prevalence estimates of medical conditions that are diagnosed based on specific cut-off points. To allow accurate cross-country comparisons of these prevalence data, harmonization of biomarker should be planned and implemented when study samples are being tested, because longitudinal studies often have changes in assay methodology or analyzer platforms over time. Since individual studies usually have limited resources for harmonization of biomarkers with other research projects, having a single reference laboratory to generate large quantities of harmonization samples with known “gold standard” values could be a more efficient approach to improve result comparability of select common assays among population-based studies in different countries.

Practitioner Points.

• Values of total cholesterol, HDL cholesterol, HbA1c, and CRP from nine nationally representative studies of the older population were highly correlated, indicating high quality of assays across all laboratories.

• There were significant differences in mean biomarker values and country ranking of at-risk levels both before and after data harmonization.

• Variability in absolute values can have a major impact on the assessment of international differences in estimates of risks for medical conditions.

Data availability statement

The equations used to construct harmonized data are provided in supplemental material. The harmonized data cannot publicly available due to privacy or ethical restrictions but they can be acquired from each of the countries.