Predictability Bounds of Electronic Health Records

Dahlem, Dominik; Maniloff, Diego; Ratti, Carlo

doi:10.1038/srep11865

Cited by 29 publications

(27 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…[89] Efforts towards creating phenotypes and discover knowledge from EHR observations must account for the biases that are inherent in EHR data due to the recording process. [4] Our ability to predict a patient's true health state from her medical history is integral to intervening in the progression of disease [10] and reducing the burden of healthcare. [90] We demonstrated that short sequences of EHR observations offer a better classification performance that using singular raw observations.…”

Section: Discussionmentioning

confidence: 99%

“…We also found that sequenced patterns carry more information about existence of the disease, and thus, are potentially also useful for predicting the disease onset before an observation of the disease is recorded in the EHRs. Dahlem et al (2015) applied a similar approach, but only to disease (ICD-9 codes) codes and showed that harnessing knowledge of disease progressions increases the predictability bounds, compared to when disease histories are treated independently. [10] Other similar work include Batal et al (2013) who used temporal abstraction and temporal pattern mining to extract the classification features.…”

Section: Discussionmentioning

confidence: 99%

“…[7][8][9] The temporal properties of EHR data signify the complexities involved in uncovering the healthcare processes and the challenges to properly incorporate temporal information in generalizable analytical techniques. [10] Despite a long tradition in informatics, translation of data mining methodologies to the complexity of the clinical workflow is slow moving. Biomedical researchers increasingly apply conventional association studies to EHR data, yet the temporality of these data has not been fully exploited.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

High-throughput Phenotyping with Temporal Sequences

Estiri

McCoy

Wagholikar

et al. 2019

Preprint

View full text Add to dashboard Cite

Electronic health records (EHRs) contain important temporal information about progression of disease and treatment outcomes. The objective of this paper is to propose and test a high-throughput approach for phenotyping using temporal sequences of EHR observations to derive predictive and interpretable data representations. Using expert labeled clinical data from a congestive heart failure (CHF) cohort, we applied a four-phase framework to mine new data representations from temporal sequences of EHR observations, reduce dimensionality of the feature space, evaluate feature importance, and classify CHF patients with CHF. We found that sequenced features significantly outperform raw EHR features in classifying heart failure patients. Compared to the gold standard created by a panel of domain experts, the best classifier using sequences of diagnosis and medication observations classified heart failure status with a receiver operating characteristic area under the curve of 0.926. Compared with their raw counterparts, the top sequenced features have lower prevalence in medical records, but present more useful information for classification as the majority of those who have the sequences are likely to have the disease. Raw EHR data are not direct reflections of true health states as they also reflect various healthcare processes. Our results demonstrated that data representations obtained from sequencing EHR observations can present novel insight about the progression of disease that is difficult to discern when clinical observation histories are treated independently. Harnessing knowledge of disease progression through temporal sequencing of EHR observations can improve computational disease classification or phenotyping.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

High-throughput Phenotyping with Temporal Sequences

Estiri

McCoy

Wagholikar

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…This was recently illustrated in a machine learning study of serum ferritin values, in which patient demographics and other concurrent laboratory results had remarkable accuracy for predicting high ferritin levels and in some cases predicted iron-deficiency anemia more accurately than measured ferritin. [37] Additional approaches such as lagged regression models[21] and other temporal-informed methods[38–41] [42] [43] can be used to account for healthcare process bias and improve the specificity and sensitivity of time-based analyses in EHR data.…”

Section: Discussionmentioning

confidence: 99%

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

“…The primary goal of this paper is to construct an automated, generalizable algorithm for computing a patient laboratory data summary that is insightful and interpretable, minimizes the information lost by parameterizing the data while minimizing or accounting for bias due to the health care process [1,2,4–6,59]. Our algorithm, the PopKLD algorithm, algorithm has seven steps and is shown in Fig.…”

Section: Methodsmentioning

confidence: 99%

Estimating summary statistics for electronic health record laboratory data for use in high-throughput phenotyping algorithms

Albers

Elhadad

Claassen

et al. 2018

Journal of Biomedical Informatics

View full text Add to dashboard Cite

We study the question of how to represent or summarize raw laboratory data taken from an electronic health record (EHR) using parametric model selection to reduce or cope with biases induced through clinical care. It has been previously demonstrated that the health care process (Hripcsak and Albers, 2012, 2013), as defined by measurement context (Hripcsak and Albers, 2013; Albers et al., 2012) and measurement patterns (Albers and Hripcsak, 2010, 2012), can influence how EHR data are distributed statistically (Kohane and Weber, 2013; Pivovarov et al., 2014). We construct an algorithm, PopKLD, which is based on information criterion model selection (Burnham and Anderson, 2002; Claeskens and Hjort, 2008), is intended to reduce and cope with health care process biases and to produce an intuitively understandable continuous summary. The PopKLD algorithm can be automated and is designed to be applicable in high-throughput settings; for example, the output of the PopKLD algorithm can be used as input for phenotyping algorithms. Moreover, we develop the PopKLD-CAT algorithm that transforms the continuous PopKLD summary into a categorical summary useful for applications that require categorical data such as topic modeling. We evaluate our methodology in two ways. First, we apply the method to laboratory data collected in two different health care contexts, primary versus intensive care. We show that the PopKLD preserves known physiologic features in the data that are lost when summarizing the data using more common laboratory data summaries such as mean and standard deviation. Second, for three disease-laboratory measurement pairs, we perform a phenotyping task: we use the PopKLD and PopKLD-CAT algorithms to define high and low values of the laboratory variable that are used for defining a disease state. We then compare the relationship between the PopKLD-CAT summary disease predictions and the same predictions using empirically estimated mean and standard deviation to a gold standard generated by clinical review of patient records. We find that the PopKLD laboratory data summary is substantially better at predicting disease state. The PopKLD or PopKLD-CAT algorithms are not meant to be used as phenotyping algorithms, but we use the phenotyping task to show what information can be gained when using a more informative laboratory data summary. In the process of evaluation our method we show that the different clinical contexts and laboratory measurements necessitate different statistical summaries. Similarly, leveraging the principle of maximum entropy we argue that while some laboratory data only have sufficient information to estimate a mean and standard deviation, other laboratory data captured in an EHR contain substantially more information than can be captured in higher-parameter models.

show abstract

Predictability Bounds of Electronic Health Records

Cited by 29 publications

References 39 publications

High-throughput Phenotyping with Temporal Sequences

High-throughput Phenotyping with Temporal Sequences

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

Estimating summary statistics for electronic health record laboratory data for use in high-throughput phenotyping algorithms

Contact Info

Product

Resources

About