Correlated errors in biased surrogates: study designs and methods for measurement error correction

Predicting type 2 diabetes mellitus (T2DM) by using phenotypic data with machine learning (ML) techniques has received significant attention in recent years. PyCaret, a low-code automated ML tool that enables the simultaneous application of 16 different algorithms, was used to predict T2DM by using phenotypic variables from the "Nurses' Health Study" and "Health Professionals' Follow-up Study" datasets. Ridge Classifier, Linear Discriminant Analysis, and Logistic Regression (LR) were the best-performing models for the male-only data subset. For the female-only data subset, LR, Gradient Boosting Classifier, and CatBoost Classifier were the strongest models. The AUC, accuracy, and precision were approximately 0.77, 0.70, and 0.70 for males and 0.79, 0.70, and 0.71 for females, respectively. The feature importance plot showed that family history of diabetes (famdb), never having smoked, and high blood pressure (hbp) were the most influential features in females, while famdb, hbp, and currently being a smoker were the major variables in males. In conclusion, PyCaret was used successfully for the prediction of T2DM by simplifying complex ML tasks. Gender differences are important to consider for T2DM prediction. Despite this comprehensive ML tool, phenotypic variables alone may not be sufficient for early T2DM prediction; genotypic variables could also be used in combination for future studies.

Road safety modelers frequently use average annual daily traffic (AADT) as a measure of exposure in regression models of expected crash frequency for road segments and intersections. Recorded AADT values at most locations are estimated by state and local transportation agencies with significant uncertainty, often by extrapolating short-term traffic counts over time and space. This uncertainty in the traffic volume estimates, often termed in a modeling context as measurement error in right-hand-side variables, can have serious effects on model estimation, including: 1) biased regression coefficient estimates; and 2) increases in dispersion. The structure and magnitude of measurement error in AADT estimates are not clearly understood by researchers or practitioners, leading to difficulties in explicitly accounting for this error in statistical road safety models, and ultimately in finding solutions for its correction. This study explores the impacts of measurement error in traffic volume estimates on statistical road safety models by employing measurement error correction approaches, including regression calibration and simulation extrapolation. The concept is demonstrated using crash, traffic, and roadway data from rural, two-lane horizontal curves in the State of Washington. The overall results show that the regression coefficient estimates with a positive coefficient were larger and those with a negative coefficient were smaller (i.e., more negative) when the measurement error correction methods were applied to the regression models of expected crash frequency. Future directions in applications of measurement error correction approaches to road safety research are provided.

In longitudinal studies, the inverse probability of treatment weighted (IPTW) method is commonly employed to estimate the effect of time-dependent treatments on an outcome of interest. However, it has been documented that when the confounders are subject to measurement error, the naive IPTW method which simply ignores measurement error leads to biased treatment effect estimation. In the existing literature, there is a lack of measurement error correction methods that fully remove measurement error effect and produce consistent treatment effect estimation. In this article, we develop a novel consistent IPTW estimation procedure for longitudinal studies. The key step of the proposed method is to use the observed data to construct a corrected function that is unbiased of the unknown IPTW function. Simulation studies reveal that the proposed method outperforms the existing consistent and approximate measurement error correction methods for IPTW estimation of the average treatment effect. Finally, we apply the proposed method to analyze a real dataset.

Metabolic energy dysregulation may affect intestinal epithelial cell homeostasis, and thus promote colorectal carcinogenesis. Yet, little is known about the association between obesity and enterocyte differentiation during tumor development.We hypothesized that an association between anthropomorphic measurements and colorectal carcinoma incidence might differ according to tumor histopathologic differentiation status. Anthropomorphic variables evaluated included Body Mass Index (BMI), Hip circumference (HC), Waist circumference (WC), and Waist‐Hip Ratio (WHR).Using data from the Nurses' Health Study and Health Professionals' Follow‐up Study, and performing duplication‐method Cox proportional hazard models, we prospectively examined an association between BMI, mean waist/hip circumference (WC/HC), and waist‐hip ratio (WHR) with the incidence of colorectal carcinoma subtypes classified by differentiation status. 120,813 participants were followed for 26 or 32 years and 1528 colorectal cancer cases with available tumor pathological data were documented.The association between BMI and colorectal cancer risk significantly differed depending on the presence or absence of poorly differentiated tumor foci (Pheterogeneity = 0.006). Higher BMI was associated with higher risk of colorectal carcinoma without poorly differentiated foci (≥30.0 vs. 18.5–22.4 kg/m2: multivariable‐adjusted hazard ratio, 1.87; 95% confidence interval, 1.49–2.34; Ptrend <0.001), but not with risk of carcinoma with poorly‐differentiated foci (Ptrend = 0.56). While in men, high WC, HC, and WHR were associated with overall incidence of colorectal carcinoma (Ptrend ≤ 0.01), there was no statistically significant association between these variables and tumor differentiation status (Pheterogeneity >0.27). Furthermore, the association between anthropomorphic variables and colorectal carcinoma did not differ significantly by overall tumor grade, mucinous differentiation, or the presence of a signet‐ring‐cell component (PHeterogeneity > 0.03, with the adjusted α of 0.01). The differential association of BMI with tumor differentiation appeared to be consistent in strata of tumor microsatellite instability and fatty acid synthase (FASN) expression, although statistical power was limited.Whilst BMI was associated with increased risk of the colorectal cancer subtype lacking poorly differentiated foci, there was no independent association between tumor differentiation and measures of central adiposity. Our findings suggest that the carcinogenic effect of generalized excess body adiposity may be stronger for tumors that retain greater degrees of intestinal differentiation.Support or Funding InformationThis work was supported by U.S. National Institutes of Health (NIH) grants (P01 CA87969 to M.J. Stampfer; UM1 CA186107 to M.J. Stampfer; P01 CA55075 to W.C. Willett; UM1 CA167552 to W.C. Willett; U01 CA167552 to W.C. Willett and L.A. Mucci; P50 CA127003 to C.S.F.; R01 CA118553 to C.S.F.; R01 CA169141 to C.S.F.; R01 CA137178 to A.T.C.; K24 DK098311 to A.T.C.; R35 CA197735 to S.O.; R01 CA151993 to S.O.; K07 CA190673 to R.N.; and K07 CA188126 to X.Z.); by Nodal Award from the Dana‐Farber Harvard Cancer Center (to S.O.); and by grants from the Project P Fund, The Friends of the Dana‐Farber Cancer Institute, Bennett Family Fund, and the Entertainment Industry Foundation through National Colorectal Cancer Research Alliance, and by Stand Up to Cancer Colorectal Cancer Dream Team Translational Research Grant (to C.S.F. and M.Gi.). J.B. was supported by the Australia Awards‐Endeavour Scholarships and Fellowships Program. K.K. was supported by grants from Overseas Research Fellowship (JP2017–775) from Japan Society for the Promotion of Science. T.H. was supported by a fellowship grant from the Mitsukoshi Health and Welfare Foundation.T.H. was supported by a fellowship grant from the Mitsukoshi Health and Welfare Foundation. L.L. was supported by a scholarship grant from Chinese Scholarship Council and a fellowship grant from Huazhong University of Science and Technology. The content is solely the responsibility of the authors and does not necessarily represent the official views of NIH. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.This abstract is from the Experimental Biology 2018 Meeting. There is no full text article associated with this abstract published in The FASEB Journal.

Cyclin D1 (CCND1) plays a key role in cell cycle control, particularly in the transition from G1 to S phase, which is regulated by cyclin-dependent kinases. A common adenine to guanine polymorphism (A870G) in the CCND1 gene has been associated with a longer-life protein and an increased risk of colorectal cancer and adenoma in some studies. Among subjects with hereditary nonpolyposis colorectal cancer, the A870G polymorphism has also been associated with a younger age of onset of colorectal cancer. We analysed 181 colorectal cancer cases and 475 matched controls and 524 adenoma cases and 517 matched controls within women in the Nurses’ Health Study (NHS) cohort, 171 colorectal cancer cases and 347 matched controls and 372 adenoma cases and 712 matched controls nested within men in the Health Professionals’ Follow-Up Study (HPFS) cohort, and 258 colorectal cancer cases and 415 matched controls within men in the Physicians’ Health Study (PHS) cohort to assess the risk associated with the CCND1 A870G genotype. Moreover, we assessed whether CCND1 genotype modified the effect of a sporadic (nonsyndromic) family history of colorectal cancer as well as the effect of other dietary and lifestyle risk factors for colorectal cancer and adenoma. In all cohorts combined, the CCND1 polymorphism did not show statistically significant associations to risk of colorectal cancer (odds ratio (OR) for A allele carriers, 1.04; 95% confidence interval (95% CI), 0.82–1.32) or adenoma (OR, 0.96; 95% CI, 0.79–1.18). The CCND1 A870G genotype was associated with a modest, although nonsignificantly elevated risk of colorectal cancer (OR, 1.59; 95% CI, 0.98–2.57) in women. In contrast, the polymorphism was not associated with increased risk of adenoma in either men or women. Among participants with the A870G genotype, a family history of colorectal cancer conferred a substantially greater risk of colorectal cancer in the women (P for interaction=0.06) and adenoma in the men (P for interaction=0.02). Current postmenopausal hormone (PMH) use was associated with a significant reduction in the risk of colorectal cancer and adenoma among women with the A870G genotype, whereas there was no effect of PMH use among those with the GG genotype. The CCND1 polymorphism appeared to confer a modest elevation in the risk of colorectal cancer among women. Moreover, the A870G genotype may enhance the protective effect of postmenopausal oestrogen use on the development of colorectal neoplasia.

We investigated whether the use of calcium channel blockers (CCBs) was associated with a reduced risk of Parkinson's disease (PD) in two large prospective cohorts: the Nurses' Health Study (NHS) and Health Professionals' Follow-Up Study (HPFS). Cox proportional hazards models were used to estimate relative risks (RRs) and 95% confidence intervals (CIs) to assess the association between use of CCBs and risk of PD adjusting for potential confounders. We identified 514 incident cases of PD during follow-up. No association between baseline use of CCBs (RR = 1.18, 95% CI: 0.73-1.92), frequency of use or duration of use of CCBs and PD risk was observed (P > 0.2 for all). These findings do not support a role for CCBs in providing neuroprotection against development of PD.

A polymorphism (G to A transition) in intron 13 of the mitochondrial enzyme monoamine oxidase B (MAOB) gene may modify, alone or by interacting with the catechol-O-methyltransferase (COMT(LL)) genotype (low enzymatic activity), the risk of idiopathic PD. Also, the association between never smoking and PD risk may be present only in people with the MAOB G allele. The authors studied two ongoing prospective cohorts-the Nurses' Health Study (121,700 women aged 30 to 55 in 1976) and the Health Professionals' Follow-up Study (51,529 men aged 40 to 75 in 1986). They identified new PD cases through 1996, selected random control subjects matched on age and study cohort, and obtained DNA samples from blood or buccal smears from 85% of the eligible cases and 84% of the control subjects. They included genotypes from 214 cases and 449 control subjects, all Caucasian. The odds ratio of PD was 1.2 (95% CI 0.9, 1.7) for MAOB genotypes G/GG/GA compared with genotypes A/AA, and 1.1 (0.7, 1.8) for COMT genotypes LL compared with HH. The odds ratio (95% CI) was 1.7 (0.7, 3.9) for those with MAOB G/GG and COMT(LL) genotypes compared with those with MAOB A/AA and COMT(HH). There was a strong association between never smoking and PD risk in all groups defined by MAOB and COMT genotypes. The findings do not support a major role of the MAOB intron 13 polymorphism in the development of PD, either by itself or by interacting with smoking.

Glaucoma is a progressive optic neuropathy and a leading cause of irreversible blindness worldwide. Primary open-angle glaucoma is the most common form, and yet the etiology of this multifactorial disease is poorly understood. We aimed to identify plasma metabolites associated with the risk of developing POAG in a case-control study (599 cases and 599 matched controls) nested within the Nurses’ Health Studies, and Health Professionals’ Follow-Up Study. Plasma metabolites were measured with LC-MS/MS at the Broad Institute (Cambridge, MA, USA); 369 metabolites from 18 metabolite classes passed quality control analyses. For comparison, in a cross-sectional study in the UK Biobank, 168 metabolites were measured in plasma samples from 2,238 prevalent glaucoma cases and 44,723 controls using NMR spectroscopy (Nightingale, Finland; version 2020). Here we show higher levels of diglycerides and triglycerides are adversely associated with glaucoma in all four cohorts, suggesting that they play an important role in glaucoma pathogenesis.

The authors evaluated the association between smoking and the incidence of psoriasis among 185,836 participants from a cohort of older women (the Nurses' Health Study, 1996-2008), a cohort of younger women (the Nurses' Health Study II, 1991-2005), and a cohort of men (Health Professionals' Follow-up Study, 1986-2006). Information on smoking was collected biennially during follow-up. The authors identified a total of 2,410 participants with incident psoriasis. Compared with never smokers, past smokers had a relative risk of incident psoriasis of 1.39 (95% confidence interval (CI): 1.27, 1.52) and current smokers had a relative risk of 1.94 (95% CI: 1.64, 2.28). For current smokers who smoked 1-14 cigarettes/day, the relative risk was 1.81 (95% CI: 1.38, 2.36); for those who smoked 15-24 cigarettes/day, the relative risk was 2.04 (95% CI: 1.68, 2.47); and for those who smoked 25 or more cigarettes/day, the relative risk was 2.29 (95% CI: 1.74, 3.01). There was a trend toward an increased risk of psoriasis with increasing pack-years or duration of smoking (P(trend) < 0.0001). The risk was highest among smokers who had 65 or more pack-years of smoking (relative risk = 2.72, 95% CI: 2.05, 3.60) and among those with a smoking duration of 30 or more years (relative risk = 1.99, 95% CI: 1.75, 2.25). The authors observed a graded reduction of risk with an increase in time since smoking cessation (P(trend) <0.0001). In this study, smoking was found to be an independent risk factor for psoriasis in both women and men. Psoriasis risk was particularly augmented for heavy smokers and persons with longer durations of smoking.

Data on the long-term association between low-carbohydrate diets and mortality are sparse. To examine the association of low-carbohydrate diets with mortality during 26 years of follow-up in women and 20 years in men. Prospective cohort study of women and men who were followed from 1980 (women) or 1986 (men) until 2006. Low-carbohydrate diets, either animal-based (emphasizing animal sources of fat and protein) or vegetable-based (emphasizing vegetable sources of fat and protein), were computed from several validated food-frequency questionnaires assessed during follow-up. Nurses' Health Study and Health Professionals' Follow-up Study. 85 168 women (aged 34 to 59 years at baseline) and 44 548 men (aged 40 to 75 years at baseline) without heart disease, cancer, or diabetes. Investigators documented 12 555 deaths (2458 cardiovascular-related and 5780 cancer-related) in women and 8678 deaths (2746 cardiovascular-related and 2960 cancer-related) in men. The overall low-carbohydrate score was associated with a modest increase in overall mortality in a pooled analysis (hazard ratio [HR] comparing extreme deciles, 1.12 [95% CI, 1.01 to 1.24]; P for trend = 0.136). The animal low-carbohydrate score was associated with higher all-cause mortality (pooled HR comparing extreme deciles, 1.23 [CI, 1.11 to 1.37]; P for trend = 0.051), cardiovascular mortality (corresponding HR, 1.14 [CI, 1.01 to 1.29]; P for trend = 0.029), and cancer mortality (corresponding HR, 1.28 [CI, 1.02 to 1.60]; P for trend = 0.089). In contrast, a higher vegetable low-carbohydrate score was associated with lower all-cause mortality (HR, 0.80 [CI, 0.75 to 0.85]; P for trend </= 0.001) and cardiovascular mortality (HR, 0.77 [CI, 0.68 to 0.87]; P for trend < 0.001). Diet and lifestyle characteristics were assessed with some degree of error. Sensitivity analyses indicated that results were probably not substantively affected by residual confounding or an unmeasured confounder. Participants were not a representative sample of the U.S. population. A low-carbohydrate diet based on animal sources was associated with higher all-cause mortality in both men and women, whereas a vegetable-based low-carbohydrate diet was associated with lower all-cause and cardiovascular disease mortality rates. National Institutes of Health.

Dietary Nicotine Intake and Risk of Parkinson Disease: A Prospective Study

Genetic susceptibility to cardiovascular disease in type 2 diabetes is influenced by abnormal metabolic status in the affected diabetic patients; and the implicated genes may differ from those in the non-diabetic populations. We conducted 3-stage genome-wide association analysis to detect the genetic factors for coronary heart disease (CHD) in existing type 2 diabetes. We analyzed 2.4 million SNPs in 1,249 diabetic women from the Nurses’ Health Study (273 CHD cases and 976 controls) and 984 diabetic men from the Health Professionals’ Follow-up Study (319 CHD cases and 665 controls), all of European ancestry. The first-stage replication was performed in Joslin Heart Study (420 CHD cases and 431 controls); and the second-stage replication was performed in Gargano Heart Study (314 CHD cases and 384 controls), and Catanzaro study (114 CHD cases and 215 controls). All the replication samples were of European ancestry and also affected by type 2 diabetes. We observed associations of genome-wide significance (the best p=2.04 * 10 -8 ) between CHD and SNPs on chromosome 1, between genes ZNF648 ( zinc finger protein 648; 1q25 .3) and GLUL (glutamate-ammonia ligase; 1q31). eQTL analysis indicates suggestive association between the associated SNPs and a potential CHD gene PPAP2B on chromosome 1 with marginal significance (p=1 * 10 -5 ). The identified SNP showed a significant association with the plasma levels of vascular cell adhesion molecule 1 ( VCAM1 ) in patients with type 2 diabetes but not in non-diabetic healthy controls. Our findings indicate that specific genetic predisposition may determine cardiovascular risk in patients with type 2 diabetes. The findings may inform the development of novel therapies for reduction of cardiovascular complications in diabetes.

Case-control designs that use population controls are compared with those that use controls selected from their relatives (i.e., siblings, cousins, or "pseudosibs" based on parental alleles) for estimating the effect of candidate genes and gene-environment interactions. The authors first evaluate the asymptotic bias in relative risk estimates resulting from using population controls when there is confounding due to population stratification. Using siblings or pseudosibs as controls completely addresses this issue, whereas cousins provide only partial protection from population stratification. Next, they show that the conventional conditional likelihood for matched case-control studies can give asymptotically biased effect estimates when applied to the pseudosib approach; the asymptotic bias is toward the null and disappears with disease rarity. They show how to reparameterize the pseudosib likelihood so this approach gives consistent effect estimates. They then show that the designs using population or pseudosib controls are generally the most efficient for estimating the main effect of a candidate gene, followed in efficiency by the design using cousins. Finally, they show that the design using sibling controls can be quite efficient when studying gene-environment interactions. In addition to asymptotic bias and efficiency issues, family-based designs might benefit from a higher motivation to participate among cases' relatives, but these designs have the disadvantage that many potential cases will be excluded from study by having no available controls.

Population pharmacokinetic (PK) data collected from routine clinical practice offers a rich source of valuable information. However, in observational population PK data, accurate time information for blood samples is often missing, resulting in measurement errors (ME) in the sampling time variable. The goal of this study was to investigate the effects on model parameters when a scheduled time is used instead of the actual blood sampling time, and to propose ME correction methods. Simulation studies were conducted based on two major factors: the curvature in PK profiles and the size of ME. As ME correction methods, transform both sides (TBS) models were developed with application of Box-Cox power transformation and Taylor expansion. The TBS models were compared to a conventional population PK model using simulations. The most important determinant of bias due to time ME was the degree of curvature (nonlinearity) in PK profiles; the smaller the curvature around sampling times, the smaller the associated bias. The second important determinant was the magnitude of ME; the larger the ME, the larger the bias. The proposed TBS models performed better than a conventional population PK modeling when curvature and ME were substantial. Time ME in sampling time can lead to bias on the parameter estimators. The following practical recommendations are provided: 1) when the curvature of PK profiles is small, conventional population PK modeling is robust to even large ME; and 2) when the curvature is moderate or large, the proposed methodology reduces bias in parameter estimates.

The derivation of the likelihood function for binary data from two types of main study/validation study designs where model covariates are measured with error is elaborated. Rather than limiting consideration to a restricted family of models with convenient mathematical properties, we suggest that empirical considerations, customized to the data at hand, should drive model choices. The joint likelihood function for the main study, in which the covariates are measured with error, and the validation study, in which they are not, is maximized, and estimation and inference proceeds using standard theory. Although the choice of the measurement error model is driven by empirical considerations, the relatively small validation study sizes typically seen may lead to misspecification, resulting in bias in estimation and inference about exposure-disease relationships. By using a nonparametric form for the measurement error model, the resulting semi-parametric methods suggested by Robins, Rotnitzky, and Zhao (1994, Journal of the American Statistical Association 89, 864-866) and Robins, Hsieh, and Newey (1995, Journal of the Royal Statistical Society, Series B 57, 409-424) are free from bias due to misspecification of the measurement error model, trading efficiency for robustness as usual. These fully and semi-parametric methods are illustrated with a detailed example from a main study/validation study of the health effects of occupational exposure to chemotherapeutics among pharmacists (Valanis et al., 1993, American Journal of Hospital Pharmacy 50, 455-462). A constant, prevalence ratio model for common binary events, with gamma covariate measurement error, is derived and empirically verified by the available data. A careful reanalysis of the data, taking measurement error fully into account, leads to a threefold increase in the log relative risk and no loss of statistical power. The semi-parametric estimates are consistent with the parametric results, providing reassurance that important bias due to misspecification of the measurement error model is unlikely.