1
|
Yuan W, Beaulieu-Jones BK, Yu KH, Lipnick SL, Palmer N, Loscalzo J, Cai T, Kohane IS. Temporal bias in case-control design: preventing reliable predictions of the future. Nat Commun 2021; 12:1107. [PMID: 33597541 PMCID: PMC7889612 DOI: 10.1038/s41467-021-21390-2] [Citation(s) in RCA: 16] [Impact Index Per Article: 5.3] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/12/2020] [Accepted: 01/22/2021] [Indexed: 02/07/2023] Open
Abstract
One of the primary tools that researchers use to predict risk is the case-control study. We identify a flaw, temporal bias, that is specific to and uniquely associated with these studies that occurs when the study period is not representative of the data that clinicians have during the diagnostic process. Temporal bias acts to undermine the validity of predictions by over-emphasizing features close to the outcome of interest. We examine the impact of temporal bias across the medical literature, and highlight examples of exaggerated effect sizes, false-negative predictions, and replication failure. Given the ubiquity and practical advantages of case-control studies, we discuss strategies for estimating the influence of and preventing temporal bias where it exists.
Collapse
Affiliation(s)
- William Yuan
- Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA.
| | | | - Kun-Hsing Yu
- Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA
| | - Scott L Lipnick
- Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA
- Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, MA, USA
- Center for Assessment Technology and Continuous Health, Massachusetts General Hospital, Boston, MA, USA
| | - Nathan Palmer
- Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA
| | - Joseph Loscalzo
- Department of Medicine, Brigham and Women's Hospital, Boston, MA, USA
| | - Tianxi Cai
- Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA
- Division of Data Sciences, VA Boston Healthcare System, Boston, MA, USA
- Department of Biostatistics, Harvard T. H. Chan School of Public Health, Boston, MA, USA
| | - Isaac S Kohane
- Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA.
| |
Collapse
|
2
|
Clouet-Foraison N, Marcovina SM, Guerra E, Aarsand AK, Coşkun A, Díaz-Garzón J, Fernandez-Calle P, Sandberg S, Ceriotti F, Carobene A. Analytical Performance Specifications for Lipoprotein(a), Apolipoprotein B-100, and Apolipoprotein A-I Using the Biological Variation Model in the EuBIVAS Population. Clin Chem 2020; 66:727-736. [PMID: 32353129 DOI: 10.1093/clinchem/hvaa054] [Citation(s) in RCA: 11] [Impact Index Per Article: 2.8] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 11/26/2019] [Accepted: 01/27/2020] [Indexed: 12/24/2022]
Abstract
BACKGROUND With increased interest in lipoprotein(a) (Lp[a]) concentration as a target for risk reduction and growing clinical evidence of its impact on cardiovascular disease (CVD) risk, rigorous analytical performance specifications (APS) and accuracy targets for Lp(a) are required. We investigated the biological variation (BV) of Lp(a), and 2 other major biomarkers of CVD, apolipoprotein A-I (apoA-I) and apolipoprotein B-100 (apoB), in the European Biological Variation Study population. METHOD Serum samples were drawn from 91 healthy individuals for 10 consecutive weeks at 6 European laboratories and analyzed in duplicate on a Roche Cobas 8000 c702. Outlier, homogeneity, and trend analysis were performed, followed by CV-ANOVA to determine BV estimates and their 95% CIs. These estimates were used to calculate APS and reference change values. For Lp(a), BV estimates were determined on normalized concentration quintiles. RESULTS Within-subject BV estimates were significantly different between sexes for Lp(a) and between women aged <50 and >50 years for apoA-I and apoB. Lp(a) APS was constant across concentration quintiles and, overall, lower than APS based on currently published data, whereas results were similar for apoA-I and apoB. CONCLUSION Using a fully Biological Variation Data Critical Appraisal Checklist (BIVAC)-compliant protocol, our study data confirm BV estimates of Lp(a) listed in the European Federation of Clinical Chemistry and Laboratory Medicine database and reinforce concerns expressed in recent articles regarding the suitability of older APS recommendations for Lp(a) measurements. Given the heterogeneity of Lp(a), more BIVAC-compliant studies on large numbers of individuals of different ethnic groups would be desirable.
Collapse
Affiliation(s)
- Noemie Clouet-Foraison
- Northwest Lipid Metabolism and Diabetes Research Laboratories, Division of Metabolism, Endocrinology, and Nutrition, University of Washington, Seattle, WA
| | - Santica M Marcovina
- Northwest Lipid Metabolism and Diabetes Research Laboratories, Division of Metabolism, Endocrinology, and Nutrition, University of Washington, Seattle, WA
| | - Elena Guerra
- Laboratory Medicine, Ospedale San Raffaele, Milan, Italy
| | - Aasne K Aarsand
- Department of Medical Biochemistry and Pharmacology, Haukeland University Hospital, Bergen, Norway.,Norwegian Organization for Quality Improvement of Laboratory Examinations (Noklus), Haraldsplass Deaconess Hospital, Bergen, Norway
| | - Abdurrahman Coşkun
- Acibadem Mehmet Ali Aydınlar University, School of Medicine, Atasehir, Istanbul, Turkey
| | - Jorge Díaz-Garzón
- Department of Laboratory Medicine, Hospital Universitario La Paz, Madrid, Spain.,Quality Analytical Commission of Spanish Society of Laboratory Medicine (SEQCML), Madrid, Spain
| | - Pilar Fernandez-Calle
- Department of Laboratory Medicine, Hospital Universitario La Paz, Madrid, Spain.,Quality Analytical Commission of Spanish Society of Laboratory Medicine (SEQCML), Madrid, Spain
| | - Sverre Sandberg
- Department of Medical Biochemistry and Pharmacology, Haukeland University Hospital, Bergen, Norway.,Norwegian Organization for Quality Improvement of Laboratory Examinations (Noklus), Haraldsplass Deaconess Hospital, Bergen, Norway.,Department of Global Public Health and Primary Care, University of Bergen, Bergen, Norway
| | - Ferruccio Ceriotti
- Central Laboratory, Fondazione IRCCS Ca' Granda, Ospedale Maggiore Policlinico, Milan, Italy
| | - Anna Carobene
- Laboratory Medicine, Ospedale San Raffaele, Milan, Italy
| |
Collapse
|
3
|
Vormittag R, Vukovich T, Stain M, Lehr S, Minar E, Pabinger I. Lipoprotein (a) in patients with spontaneous venous thromboembolism. Thromb Res 2007; 120:15-20. [PMID: 16643992 DOI: 10.1016/j.thromres.2006.03.002] [Citation(s) in RCA: 18] [Impact Index Per Article: 1.1] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/06/2005] [Revised: 02/24/2006] [Accepted: 03/07/2006] [Indexed: 10/24/2022]
Abstract
INTRODUCTION Elevated lipoprotein (a) (Lp (a)) has been established as a risk factor of coronary heart disease and stroke. Findings concerning the risk of venous thromboembolism (VTE) in adults are contradictory. The aim of our study was to investigate, whether elevated Lp (a) levels are an independent risk factor of spontaneous symptomatic venous thromboembolism (VTE). Our study was further designed to detect differences in risk profiles between thrombosis patients with and without symptomatic PE. MATERIALS AND METHODS We investigated Lp (a) in 128 patients with spontaneous symptomatic deep vein thrombosis (DVT, group 1), 105 with spontaneous symptomatic pulmonary embolism with or without DVT (PE, group 2) and 122 healthy controls. Lp (a) was measured with an immunoturbidimetric assay (Tina-quant(R), Roche, Grenzach-Wyhlen, Germany) on a Hitachi-Modular system. RESULTS Lp (a) levels (mg/L) were not significantly different among groups, median levels (25th-75th percentiles) were 170 (51-386) in group 1, 140 (<20-427) in group 2 and 126 (54-331) in controls, respectively. As continuous variable, odds ratios for VTE for a 100 mg/L increase of Lp (a) were 1.1 [95% confidence interval 0.98-1.2] for group 1 versus controls and 1.1 [0.95-1.2] for group 2 versus controls. The prevalence of Lp (a) above 300 mg/L was not significantly different among patients and controls (group 1: 30%, group 2: 32% and controls: 25%, p=0.4, p=0.2, respectively). CONCLUSIONS In conclusion we found no association between Lp (a) and VTE regardless whether DVT occurred together with PE or not.
Collapse
Affiliation(s)
- Rainer Vormittag
- Department of Internal Medicine I, Division of Hematology and Blood Coagulation, University Hospital of Vienna, Austria
| | | | | | | | | | | |
Collapse
|
4
|
Hernández C, Francisco G, Chacón P, Mesa J, Simó R. Biological variation of lipoprotein(a) in a diabetic population. Analysis of the causes and clinical implications. Clin Chem Lab Med 2003; 41:1075-80. [PMID: 12964817 DOI: 10.1515/cclm.2003.166] [Citation(s) in RCA: 7] [Impact Index Per Article: 0.3] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 12/19/2022]
Abstract
The aims of the present study were to evaluate the biological variability of lipoprotein(a) (Lp(a)) in diabetic patients and to investigate the biological sources of this variability. Lp(a) was measured by ELISA in four serum specimens collected in 3-month intervals from 70 patients. The other parameters analyzed were: total cholesterol, high density lipoprotein-cholesterol (HDL-C), low density lipoprotein-cholesterol (LDL-C), triglycerides, glucose, HbA and albumin excretion rate. The overall biological within-subject variance (CVb) was 31.7%, and it was inversely correlated with Lp(a) serum levels. According to the initial ranges of Lp(a) serum levels (< 15, 15-30 and > 30 mg/dl) the CVb were 42.3%, 24.1% and 23.7%, respectively. In multivariate analysis the total intra-individual coefficient of variation (CVt) of triglycerides and the CVt of the albumin excretion rate (AER) were independently associated with the CVb of Lp(a) (R2 = 0.54). The intra-individual biological variation of Lp(a) produced a misclassification of 20% of diabetic patients for cardiovascular risk attributable to this lipoprotein. In conclusion, the higher biological variability of Lp(a) observed in diabetic patients suggests that a single determination could be inaccurate to assess the cardiovascular risk associated with this lipoprotein, at least in those patients in whom serum levels are near the cut-off considered as risk for cardiovascular disease (> 30 mg/dl). Finally, triglycerides and AER are the main factors influencing Lp(a) serum levels in the diabetic population.
Collapse
Affiliation(s)
- Cristina Hernández
- Diabetes Research Unit, Hospital General Vall d'Hebron, Barcelona, Spain
| | | | | | | | | |
Collapse
|
5
|
Alaoui M, Merle C, Garnotel R, Jolly D, Durlach V, Gillery P. Study of apo(a) length polymorphism and lipoprotein(a) concentrations in subjects with single or double apo(a) isoforms. Clin Chem Lab Med 2003; 41:634-9. [PMID: 12812260 DOI: 10.1515/cclm.2003.096] [Citation(s) in RCA: 2] [Impact Index Per Article: 0.1] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/15/2022]
Abstract
Cardiovascular risk is associated with high lipoprotein(a) (Lp(a)) concentrations and low molecular weight apolipoprotein(a) (apo(a)) isoforms. We studied the relationship between these two biological parameters, particularly in subjects expressing two apo(a) isoforms. Plasma Lp(a) was measured by immunonephelometry in 530 unrelated Caucasian patients at high cardiovascular risk, and apo(a) size determined by immunoblotting using a recombinant standard. Two, one, or no apo(a) isoforms were detected in 258, 270, and 2 subjects, respectively. Lp(a) concentrations showed a non-Gaussian distribution, being higher in the 'double band' than in the 'single band' group (median 0.42 vs. 0.11 g/l, p < 0.0005). Apo(a) size distribution was bimodal, with two frequency peaks at 18 kringles (K) and 27 K. Small size apo(a) isoforms were more frequently found in the 'double band' group, where major isoforms were of lower size than minor isoforms (median 20 vs. 27 K). Regression analysis showed that apo(a) gene length accounted for 33% of Lp(a) variation, with a threshold effect at 20 K, no correlation being found over this value. The minor apo(a) isoform did not significantly influence Lp(a) concentration. These data confirm the relationship between apo(a) size and Lp(a) concentration and suggest that the assessment of cardiovascular risk should take into account the threshold effect at 20 K and the absence of influence of the minor apo(a) isoform.
Collapse
Affiliation(s)
- Mariam Alaoui
- Laboratory of Biochemistry, Robert Debré Hospital, CHU of Reims, Reims, France
| | | | | | | | | | | |
Collapse
|