Biases introduced by filtering electronic health records for patients with “complete data”

Weber, Griffin M; Adams, William; Bernstam, Elmer V.; Bickel, Jonathan; Fox, Kathe; Marsolo, Keith; Raghavan, V.; Turchin, Alexander; Zhou, Xiaobo; Murphy, Shawn N.; Mandl, Kenneth D.

doi:10.1093/jamia/ocx071

Cited by 72 publications

(53 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[11–14, 17–19] It is known that invoking certain data sufficiency requirements selects for a sicker patient population, since these patients have more frequent data collection[14, 17]. In our cohort, shorter time between dates of lipid and BP measurements was associated with a greater number of comorbidities and with poorer BP and cholesterol values.…”

Section: Discussionmentioning

confidence: 94%

“…In the analytic phase of research, investigators often limit analyses only to patients with a complete set of data. [14, 31] For example, the eMERGE network has defined completeness as having one biobanked sample, at least two clinical visits, and data from each of several data categories. [36] Our results and those of others suggest that such data sufficiency requirements might introduce bias that should be examined before investigators choose this strategy over other methods to account for missing data,[14] including multiple imputation.…”

Section: Discussionmentioning

confidence: 99%

“…[10] As a result, the content and frequency of data collection that is appropriate for these primary uses can introduce bias when the data are used for research. [11–14] Bias exists when the association observed between an exposure and outcome is not entirely due to causality[15] and can threaten the internal validity of a study’s findings. Although dozens of biases have been described, most can be assigned to one of three categories: selection bias, information bias, and confounding.…”

Section: Introductionmentioning

confidence: 99%

“…Limiting the analyses only to those patients with complete baseline data at one timepoint might decrease the yield of eligible patients and introduce selection bias, error in cohort selection that favors patients who access care most often. [14, 15, 18] However, expanding the definition of the baseline timeframe might introduce information bias, error in cohort characterization,[16, 18] if patients with less frequent contact are more likely to have inaccurate measurements of their risk factors at any one instance. Together, these errors that result from the way data are collected in EHR have been termed healthcare process bias.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Yield and bias in defining a cohort study baseline from electronic health record data

Vassy

Cho

et al. 2018

Journal of Biomedical Informatics

View full text Add to dashboard Cite

Aims Despite growing interest in using electronic health records (EHR) to create longitudinal cohort studies, the distribution and missingness of EHR data might introduce selection bias and information bias to such analyses. We aimed to examine the yield and potential for these healthcare process biases in defining a study baseline using EHR data, using the example of cholesterol and blood pressure (BP) measurements. Methods We created a virtual cohort study of cardiovascular disease (CVD) from patients with eligible cholesterol profiles in the New England (NE) and Southeast (SE) networks of the Veterans Health Administration in the United States. Using clinical data from the EHR, we plotted the yield of patients with BP measurements within an expanding timeframe around an index date of cholesterol testing. We compared three groups: 1) patients with BP from the exact index date; 2) patients with BP not on the index date but within the network-specific 90th percentile around the index date; and 3) patients with no BP within the network-specific 90th percentile. Results Among 589,361 total patients in the two networks, 146,636 (61.0%) of 240,479 patients from NE and 289,906 (83.1%) of 348,882 patients from SE had BP measurements on the index date. Ninety percent had BP measured within 11 days of the index date in NE and within 5 days of the index date in SE. Group 3 in both networks had fewer available race data, fewer comorbidities and CVD medications, and fewer health system encounters. Conclusions Requiring same-day risk factor measurement in the creation of a virtual CVD cohort study from EHR data might exclude 40% of eligible patients, but including patients with infrequent visits might introduce bias. Data visualization can inform study-specific strategies to address these challenges for the research use of EHR data.

show abstract

Section: Discussionmentioning

confidence: 94%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Yield and bias in defining a cohort study baseline from electronic health record data

Vassy

Cho

et al. 2018

Journal of Biomedical Informatics

View full text Add to dashboard Cite

show abstract

“…To tune hyperparameters if any in these models, we mask out 20% observed measurements in the training set as a validation set and tune hyperparameters on the validation set. We introduce π (1) , π (2) and π (3) as hyperparameters of our imputation model. In our experiments, all mixture models start with the same mixing weights, i.e., π (k) v,b = π (k) for k = 1, 2, 3.…”

Section: Evaluation Of Imputation Qualitymentioning

confidence: 99%

Mixture-based Multiple Imputation Model for Clinical Data with a Temporal Dimension

Xue¹,

Klabjan²,

Luo³

2019

2019 IEEE International Conference on Big Data (Big Data)

View full text Add to dashboard Cite

The problem of missing values in multivariable time series is a key challenge in many applications such as clinical data mining. Although many imputation methods show their effectiveness in many applications, few of them are designed to accommodate clinical multivariable time series. In this work, we propose a multiple imputation model that capture both crosssectional information and temporal correlations. We integrate Gaussian processes with mixture models and introduce individualized mixing weights to handle the variance of predictive confidence of Gaussian process models. The proposed model is compared with several state-of-the-art imputation algorithms on both real-world and synthetic datasets. Experiments show that our best model can provide more accurate imputation than the benchmarks on all of our datasets.

show abstract

Second-Line Pharmaceutical Treatments for Patients with Type 2 Diabetes

Vashisht,

Patel,

Dahm

et al. 2023

JAMA Netw Open

View full text Add to dashboard Cite

ImportanceAssessing the relative effectiveness and safety of additional treatments when metformin monotherapy is insufficient remains a limiting factor in improving treatment choices in type 2 diabetes.ObjectiveTo determine whether data from electronic health records across the University of California Health system could be used to assess the comparative effectiveness and safety associated with 4 treatments in diabetes when added to metformin monotherapy.Design, Setting, and ParticipantsThis multicenter, new user, multidimensional propensity score–matched retrospective cohort study with leave-one-medical-center-out (LOMCO) sensitivity analysis used principles of emulating target trial. Participants included patients with diabetes receiving metformin who were then additionally prescribed either a sulfonylurea, dipeptidyl peptidase-4 inhibitor (DPP4I), sodium-glucose cotransporter-2 inhibitor (SGLT2I), or glucagon-like peptide-1 receptor agonist (GLP1RA) for the first time and followed-up over a 5-year monitoring period. Data were analyzed between January 2022 and April 2023.ExposureTreatment with sulfonylurea, DPP4I, SGLT2I, or GLP1RA added to metformin monotherapy.Main Outcomes and MeasuresThe main effectiveness outcome was the ability of patients to maintain glycemic control, represented as time to metabolic failure (hemoglobin A1c [HbA1c] ≥7.0%). A secondary effectiveness outcome was assessed by monitoring time to new incidence of any of 28 adverse outcomes, including diabetes-related complications while treated with the assigned drug. Sensitivity analysis included LOMCO.ResultsThis cohort study included 31 852 patients (16 635 [52.2%] male; mean [SD] age, 61.4 [12.6] years) who were new users of diabetes treatments added on to metformin monotherapy. Compared with sulfonylurea in random-effect meta-analysis, treatment with SGLT2I (summary hazard ratio [sHR], 0.75 [95% CI, 0.69-0.83]; I2 = 37.5%), DPP4I (sHR, 0.79 [95% CI, 0.75-0.84]; I2 = 0%), GLP1RA (sHR, 0.62 [95% CI, 0.57-0.68]; I2 = 23.6%) were effective in glycemic control; findings from LOMCO sensitivity analysis were similar. Treatment with SGLT2I showed no significant difference in effectiveness compared with GLP1RA (sHR, 1.26 [95% CI, 1.12-1.42]; I2 = 47.3%; no LOMCO) or DPP4I (sHR, 0.97 [95% CI, 0.90-1.04]; I2 = 0%). Patients treated with DPP4I and SGLT2I had fewer cardiovascular events compared with those treated with sulfonylurea (DPP4I: sHR, 0.84 [95% CI, 0.74-0.96]; I2 = 0%; SGLT2I: sHR, 0.78 [95% CI, 0.62-0.98]; I2 = 0%). Patients treated with a GLP1RA or SGLT2I were less likely to develop chronic kidney disease (GLP1RA: sHR, 0.75 [95% CI 0.6-0.94]; I2 = 0%; SGLT2I: sHR, 0.77 [95% CI, 0.61-0.97]; I2 = 0%), kidney failure (GLP1RA: sHR, 0.69 [95% CI, 0.56-0.86]; I2 = 9.1%; SGLT2I: sHR, 0.72 [95% CI, 0.59-0.88]; I2 = 0%), or hypertension (GLP1RA: sHR, 0.82 [95% CI, 0.68-0.97]; I2 = 0%; SGLT2I: sHR, 0.73 [95% CI, 0.58-0.92]; I2 = 38.5%) compared with those treated with a sulfonylurea. Patients treated with an SGLT2I, vs a DPP4I, GLP1RA, or sulfonylurea, were less likely to develop indicators of chronic hepatic dysfunction (sHR vs DPP4I, 0.68 [95% CI, 0.49-0.95]; I2 = 0%; sHR vs GLP1RA, 0.66 [95% CI, 0.48-0.91]; I2 = 0%; sHR vs sulfonylurea, 0.60 [95% CI, 0.44-0.81]; I2 = 0%), and those treated with a DPP4I were less likely to develop new incidence of hypoglycemia (sHR, 0.48 [95% CI, 0.36-0.65]; I2 = 22.7%) compared with those treated with a sulfonylurea.Conclusions and RelevanceThese findings highlight familiar medication patterns, including those mirroring randomized clinical trials, as well as providing new insights underscoring the value of robust clinical data analytics in swiftly generating evidence to help guide treatment choices in diabetes.

show abstract

Biases introduced by filtering electronic health records for patients with “complete data”

Abstract: As investigators design studies, they need to balance their confidence in the completeness of the data with the effects of placing requirements on the data on the resulting patient cohort.

Cited by 72 publications

References 20 publications

Yield and bias in defining a cohort study baseline from electronic health record data

Yield and bias in defining a cohort study baseline from electronic health record data

Mixture-based Multiple Imputation Model for Clinical Data with a Temporal Dimension

Second-Line Pharmaceutical Treatments for Patients with Type 2 Diabetes

Contact Info

Product

Resources

About