Assessing occurrence of hypoglycemia and its severity from electronic health records of patients with type 2 diabetes mellitus

Validation of a natural language processing algorithm using national reporting data to improve identification of anesthesia-related ADVerse evENTs: The "ADVENTURE" study

BACKGROUND

Reporting and analysis of adverse events (AE) is associated with improved health system learning, quality outcomes, and patient safety. Manual text analysis is time-consuming, costly, and prone to human errors. We aimed to demonstrate the feasibility of novel machine learning and natural language processing (NLP) approaches for early predictions of adverse events and provide input to direct quality improvement and patient safety initiatives.

METHODS

We used machine learning to analyze 9559 continuously reported AE by clinicians and healthcare systems to the French National Health accreditor (HAS) between January 1, 2009, and December 31, 2020 . We validated the labeling of 135,000 unique de-identified AE reports and determined the associations between different system's root causes and patient consequences. The model was validated by independent expert anesthesiologists.

RESULTS

The machine learning (ML) and Artificial Intelligence (AI) model trained on 9559 AE datasets accurately categorized 8800 (88%) of reported AE. The three most frequent AE types were "difficult orotracheal intubation" (16.9% of AE reports), "medication error" (10.5%), and "post-induction hypotension" (6.9%). The accuracy of the AI model reached 70.9% sensitivity, 96.6% specificity for "difficult intubation", 43.2% sensitivity, and 98.9% specificity for "medication error."

CONCLUSIONS

This unsupervised ML method provides an accurate, automated, AI-supported search algorithm that ranks and helps to understand complex risk patterns and has greater speed, precision, and clarity when compared to manual human data extraction. Machine learning and Natural language processing (NLP) models can effectively be used to process natural language AE reports and augment expert clinician input. This model can support clinical applications and methodological standards and used to better inform and enhance decision-making for improved risk management and patient safety.

TRIAL REGISTRATION

The study was approved by the ethics committee of the French Society of Anesthesiology (IRB 00010254-2020-20) and the CNIL (CNIL: 118 58 95) and the study was registered with ClinicalTrials.gov (NCT: NCT05185479).

Deep-learning-based natural-language-processing models to identify cardiovascular disease hospitalisations of patients with diabetes from routine visits' text

Writing notes is the most widespread method to report clinical events. Therefore, most of the information about the disease history of a patient remains locked behind free-form text. Natural language processing (NLP) provides a solution to automatically transform free-form text into structured data. In the present work, electronic healthcare records data of patients with diabetes were used to develop deep-learning based NLP models to automatically identify, within free-form text describing routine visits, the occurrence of hospitalisations related to cardiovascular disease (CVDs), an outcome of diabetes. Four possible time windows of increasing level of expected difficulty were considered: infinite, 24 months, 12 months, and 6 months. Model performance was evaluated by means of the area under the precision recall curve, as well as precision, recall, and F1-score after thresholding. Results showed that the proposed NLP approach was successful for both the infinite and 24-month windows, while, as expected, performance deteriorated with shorter time windows. Possible clinical applications of tools based on the proposed NLP approach include the retrospective filling of medical records with respect to a patient's CVD history for epidemiological and research purposes as well as for clinical decision making.

iGlarLixi versus basal plus Rapid-Acting insulin in adults with type 2 diabetes advancing from basal insulin therapy: The SoliSimplify Real-World study

AIM

For people with suboptimally controlled type 2 diabetes (T2D) on basal insulin (BI), guidelines recommend several treatment advancement options. This study compared the clinical effectiveness of once-daily iGlarLixi versus a multiple-injection BI + rapid acting insulin (RAI) regimen in adults with T2D advancing from BI therapy in real-world clinical practice.

MATERIALS AND METHODS

Electronic medical records from the Observational Medical Outcomes Partnership (OMOP) database were analysed retrospectively using propensity score matching to compare therapy advancement with iGlarLixi or BI + RAI in US adults ≥18 years with T2D on BI who had ≥1 valid glycated haemoglobin (HbA1c) value at baseline and at the 6-month follow-up. The primary objective was non-inferiority of iGlarLixi to BI + RAI in HbA1c change from baseline to 6 months (margin 0.3%).

RESULTS

Propensity score matching generated cohorts with balanced baseline characteristics (N = 814 in each group). HbA1c reduction from baseline to 6 months with iGlarLixi was non-inferior to BI + RAI [mean difference (95% confidence interval): 0.1 (-0.1, 0.2)%; one-sided p = .0032]. At 6 months, weight gain was significantly lower with iGlarLixi than with BI + RAI [-0.8 (-1.3, -0.2) kg; two-sided p = .0069]. Achievement of HbA1c <7% without hypoglycaemia and weight gain were similar between groups [odds ratio (95% confidence interval): 1.15 (0.81, 1.63); p = .4280]. Hypoglycaemia was low in both groups, probably because of underreporting.

CONCLUSIONS

In real-world clinical practice, glycaemic outcomes 6 months after treatment advancement from BI are similar for people with T2D using iGlarLixi versus BI + RAI, with iGlarLixi leading to less weight gain.

Novel methodology for the evaluation of symptoms reported by patients with newly diagnosed atrial fibrillation: Application of natural language processing to electronic medical records data

INTRODUCTION

Understanding symptom patterns in atrial fibrillation (AF) can help in disease management. We report on the application of natural language processing (NLP) to electronic medical records (EMRs) to capture symptom reports in patients with newly diagnosed (incident) AF.

METHODS AND RESULTS

This observational retrospective study included adult patients with an index diagnosis of incident AF during January 1, 2016 through June 30, 2018, in the Optum datasets. The baseline and follow-up periods were 1 year before/after the index date, respectively. The primary objective was identification of the following predefined symptom reports: dyspnea or shortness of breath; syncope, presyncope, lightheadedness, or dizziness; chest pain; fatigue; and palpitations. In an exploratory analysis, the incidence rates of symptom reports and cardiovascular hospitalization were assessed in propensity-matched patient cohorts with incident AF receiving first-line dronedarone or sotalol. Among 30 447 patients with an index AF diagnosis, the NLP algorithm identified at least 1 predefined symptom in 9734 (31.9%) patients. The incidence rate of symptom reports was highest at 0-3 months post-diagnosis and lower at >3-6 and >6-12 months (pre-defined timepoints). Across all time periods, the most common symptoms were dyspnea or shortness of breath, followed by syncope, presyncope, lightheadedness, or dizziness. Similar temporal patterns of symptom reports were observed among patients with prescriptions for dronedarone or sotalol as first-line treatment.

CONCLUSION

This study illustrates that NLP can be applied to EMR data to characterize symptom reports in patients with incident AF, and the potential for these methods to inform comparative effectiveness.

Potential Risk of Overtreatment in Patients with Type 2 Diabetes Aged 75 Years or Older: Data from a Population Database in Catalonia, Spain

Aim: To assess the potential risk of overtreatment in patients with type 2 diabetes (T2DM) aged 75 years or older in primary care. Methods: Electronic health records retrieved from the SIDIAP database (Catalonia, Spain) in 2016. Variables: age, gender, body mass index, registered hypoglycemia, last HbA1c and glomerular filtration rates, and prescriptions for antidiabetic drugs. Potential overtreatment was defined as having HbA1c < 7% or HbA1c < 6.5% in older patients treated with insulin, sulfonylureas, or glinides. Results: From a total population of 138,374 T2DM patients aged 75 years or older, 123,515 had at least one HbA1c available. An HbA1c below 7.0% was present in 59.1% of patients, and below 6.5% in 37.7%. Overall, 23.0% of patients were treated with insulin, 17.8% with sulfonylureas, and 6.6% with glinides. Potential overtreatment (HbA1c < 7%) was suspected in 26.6% of patients treated with any high-risk drug, 47.8% with sulfonylureas, 43.5% with glinides, and 28.1% with insulin. Using the threshold of HbA1c < 6.5%, these figures were: 21.6%, 24.4%, 17.9%, and 12.3%, respectively. Conclusion: One in four older adults with T2DM treated with antidiabetic drugs associated with a high risk of hypoglycemia might be at risk of overtreatment. This risk is higher in those treated with sulfonylureas or glinides than with insulin.

Best Practices on Big Data Analytics to Address Sex-Specific Biases in Our Understanding of the Etiology, Diagnosis, and Prognosis of Diseases

A bias in health research to favor understanding diseases as they present in men can have a grave impact on the health of women. This paper reports on a conceptual review of the literature on machine learning or natural language processing (NLP) techniques to interrogate big data for identifying sex-specific health disparities. We searched Ovid MEDLINE, Embase, and PsycINFO in October 2021 using synonyms and indexing terms for (a) "women," "men," or "sex"; (b) "big data," "artificial intelligence," or "NLP"; and (c) "disparities" or "differences." From 902 records, 22 studies met the inclusion criteria and were analyzed. Results demonstrate that the inclusion by sex is inconsistent and often unreported, although the inclusion of men in these studies is disproportionately less than women. Even though artificial intelligence and NLP techniques are widely applied in health research, few studies use them to take advantage of unstructured text to investigate sex-related differences or disparities. Researchers are increasingly aware of sex-based data bias, but the process toward correction is slow. We reflect on best practices on using big data analytics to address sex-specific biases in understanding the etiology, diagnosis, and prognosis of diseases. Expected final online publication date for the Annual Review of Biomedical Data Science, Volume 5 is August 2022. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.

Heavy head lice infestation in an adolescent girl following benzodiazepine poisoning

Loss of consciousness (LOC) is one of the most common causes of emergency department (ED) visits. It may be due to intoxication or hypoglycemia. We present a 15‐year‐old girl who was referred with heavy head lice and LOC to the ED in the north of Iran.

Heavy head lice were detected on a 15‐year‐old girl who was admitted to the emergency ward with loss of consciousness and hypoglycemia following benzodiazepine poisoning.

Retrospective Observational Real-World Outcome Study to Evaluate Safety Among Patients With Erectile Dysfunction (ED) With Co-Possession of Tadalafil and Anti-Hypertensive Medications (anti-HTN)

BACKGROUND

Erectile dysfunction (ED) is a common condition affecting male adults and may be associated with hypertension, diabetes, hyperlipidemia, and obesity. Phosphodiesterase type 5 (PDE5) inhibitors, such as tadalafil, are the first-line drug therapy for ED. Studies and the current prescribing information of these PDE5 inhibitors indicate they are mechanistic mild vasodilators and, as such, concomitant use of a PDE5 inhibitor with anti-hypertensive medication may lead to drops in blood pressure due to possible drug-drug interaction.

AIM

Evaluate risks of hypotensive/cardiovascular outcomes in a large cohort of patients with ED that have co-possession of prescriptions for tadalafil and hypertensive medications versus either medication/s alone.

METHODS

A cohort study conducted within an electronic health record database (Optum) representing hospitals across the US. Adult male patients prescribed tadalafil and/or anti-hypertensive medications from January 2012 to December 2017 were eligible. Possession periods were defined by the time patients likely had possession of medication, with propensity score-matched groups used for comparison.

OUTCOMES

Risk of hypotensive/cardiovascular outcomes were measured using diagnostic codes and NLP algorithms during possession periods of tadalafil + anti-hypertensive versus either medication/s alone.

RESULTS

In total there were 127,849 tadalafil + anti-hypertensive medication possession periods, 821,359 anti-hypertensive only medication possession periods, and 98,638 tadalafil only medication possession periods during the study; 126,120 were successfully matched. Adjusted-matched incidence rate ratios (IRRs) for the anti-hypertensive only possession periods compared with tadalafil + anti-hypertensive periods of diagnosed outcomes were all below 1. Two outcomes had a 95% confidence interval (CI) that did not include 1.0: ventricular arrhythmia (IRR 0.79; 95% CI 0.66, 0.94) and diagnosis of hypotension (IRR 0.79; 95% CI 0.71, 0.89).

CLINICAL IMPLICATIONS

Provides real world evidence that co-possession of tadalafil and anti-hypertensive medications does not increase risk of hypotensive/cardiovascular outcomes beyond that observed for patients in possession of anti-hypertensive medications only.

STRENGTHS AND LIMITATIONS

EHR data are valuable for the evaluation of real world outcomes, however, the data are retrospective and collected for clinical patient management rather than research. Prescription data represent the intent of the prescriber and not use by the patient. Residual bias cannot be ruled out, despite propensity score matching, due to unobserved patient characteristics and severity that are not fully reflected in the EHR database.

CONCLUSION

In the studied real world patients, this study did not demonstrate an increased risk of hypotensive or cardiovascular outcomes associated with co-possession of tadalafil and anti-hypertensive medications beyond that observed for patients in possession of anti-hypertensive medications only. Nunes AP, Seeger JD, Stewart A, et al., Retrospective Observational Real-World Outcome Study to Evaluate Safety Among Patients With Erectile Dysfunction (ED) With Co-Possession of Tadalafil and Anti-Hypertensive Medications (anti-HTN). J Sex Med 2022;19:74-82.

Identifying Patients with Hypoglycemia Using Natural Language Processing: A Systematic Literature Review (Preprint)

Background

Accurately identifying patients with hypoglycemia is key to preventing adverse events and mortality. Natural language processing (NLP), a form of artificial intelligence, uses computational algorithms to extract information from text data. NLP is a scalable, efficient, and quick method to extract hypoglycemia-related information when using electronic health record data sources from a large population.

Objective

The objective of this systematic review was to synthesize the literature on the application of NLP to extract hypoglycemia from electronic health record clinical notes.

Methods

Literature searches were conducted electronically in PubMed, Web of Science Core Collection, CINAHL (EBSCO), PsycINFO (Ovid), IEEE Xplore, Google Scholar, and ACL Anthology. Keywords included hypoglycemia, low blood glucose, NLP, and machine learning. Inclusion criteria included studies that applied NLP to identify hypoglycemia, reported the outcomes related to hypoglycemia, and were published in English as full papers.

Results

This review (n=8 studies) revealed heterogeneity of the reported results related to hypoglycemia. Of the 8 included studies, 4 (50%) reported that the prevalence rate of any level of hypoglycemia was 3.4% to 46.2%. The use of NLP to analyze clinical notes improved the capture of undocumented or missed hypoglycemic events using International Classification of Diseases, Ninth Revision (ICD-9), and International Classification of Diseases, Tenth Revision (ICD-10), and laboratory testing. The combination of NLP and ICD-9 or ICD-10 codes significantly increased the identification of hypoglycemic events compared with individual methods; for example, the prevalence rates of hypoglycemia were 12.4% for International Classification of Diseases codes, 25.1% for an NLP algorithm, and 32.2% for combined algorithms. All the reviewed studies applied rule-based NLP algorithms to identify hypoglycemia.

Conclusions

The findings provided evidence that the application of NLP to analyze clinical notes improved the capture of hypoglycemic events, particularly when combined with the ICD-9 or ICD-10 codes and laboratory testing.

Insulin glargine 300 units/mL for the treatment of individuals with type 2 diabetes in the real world: A review of the DELIVER programme

Evidence from randomized controlled trials (RCTs) has shown that second-generation basal insulin (BI) analogues, insulin glargine 300 U/mL (Gla-300) and insulin degludec (IDeg), provide similar glycaemic control, with a lower risk of hypoglycaemia compared with the first-generation BI analogue insulin glargine 100 U/mL (Gla-100) in people with type 2 diabetes (T2D). However, the highly selected participants and frequent follow-up of RCTs may not be truly representative of real-life clinical practice. It is important to assess the safety and effectiveness of these second-generation BI analogues in real-life clinical practice settings. The DELIVER programme utilized electronic healthcare records from the United States to compare clinical outcomes in people with T2D who received either Gla-300 or other BI analogues in real-world clinical practice. This review provides a concise overview of the results of the DELIVER studies. Overall, Gla-300 provided similar antihyperglycaemic effectiveness and a lower risk of hypoglycaemia versus the first-generation BI analogues Gla-100 and insulin detemir in people with T2D who had switched BIs. In those who were insulin-naïve, initiation with Gla-300 versus Gla-100 was associated with significantly better antihyperglycaemic effectiveness and similar or lower hypoglycaemic risk. Both glycaemic control and hypoglycaemia risk were also shown to be similar with Gla-300 and IDeg, in people who had switched BIs and in those who were insulin-naïve. In addition, the DELIVER 2 study reported that people with T2D who switched to Gla-300 had reduced healthcare resource utilization, with an overall saving of US$1439 per person per year compared with those who switched to another BI analogue. Overall, the real-world DELIVER programme showed that the glycaemic control with a low risk of hypoglycaemia observed with Gla-300 in RCTs was also seen in standard clinical practice.

Association of Social Behaviors With Community Discharge in Patients with Total Hip and Knee Replacement

OBJECTIVES

Understand the association between social determinants of health and community discharge after elective total joint arthroplasty.

DESIGN

Retrospective cohort design using Optum de-identified electronic health record dataset.

SETTING AND PARTICIPANTS

A total of 38 hospital networks and 18 non-network hospitals in the United States; 79,725 patients with total hip arthroplasty and 136,070 patients with total knee arthroplasty between 2011 and 2018.

METHODS

Logistic regression models were used to examine the association among pain, weight status, smoking status, alcohol use, substance disorder, and postsurgical community discharge, adjusted for patient demographics.

RESULTS

Mean ages for patients with hip and knee arthroplasty were 64.5 (SD 11.3) and 65.9 (SD 9.6) years; most patients were women (53.6%, 60.2%), respectively. The unadjusted community discharge rate was 82.8% after hip and 81.1% after knee arthroplasty. After adjusting for demographics, clinical factors, and behavioral factors, we found obesity [hip: odds ratio (OR) 0.81, 95% confidence interval (CI) 0.76-0.85; knee: OR 0.73, 95% CI 0.69-0.77], current smoking (hip: OR 0.82, 95% CI 0.77-0.88; knee: OR 0.90, 95% CI 0.85-0.95), and history of substance use disorder (hip: OR 0.55, 95% CI 0.50-0.60; knee: OR 0.57, 95% CI 0.53-0.62) were associated with lower odds of community discharge after hip and knee arthroplasty, respectively.

CONCLUSIONS AND IMPLICATIONS

Social determinants of health are associated with odds of community discharge after total hip and knee joint arthroplasty. Our findings demonstrate the value of using electronic health record data to analyze more granular patient factors associated with patient discharge location after total joint arthroplasty. Although bundled payment is increasing community discharge rates, post-acute care facilities must be prepared to manage more complex patients because odds of community discharge are diminished in those who are obese, smoking, or have a history of substance use disorder.

Sex disparities in patients with symptomatic severe aortic stenosis

BACKGROUND

We evaluated whether there is equitable distribution across sexes of treatment and outcomes for aortic valve replacement (AVR), via surgical (SAVR) or transcatheter (TAVR) methods, in symptomatic severe aortic stenosis (ssAS) patients.

METHODS

Using de-identified data, we identified 43,822 patients with ssAS (2008-2016). Multivariate competing risk models were used to determine the likelihood of any AVR, while accounting for the competing risk of death. Association between sex and 1-year mortality, stratified by AVR status, was evaluated using multivariate Cox regression models with AVR as a time-dependent variable.

RESULTS

Among patients with ssAS, 20,986 (47.9%) were female. Females were older (median age 81 vs. 78, P<0.001), more likely to have body mass index <20 (8.5% vs. 3.5%), and home oxygen use (4.4% vs. 3.4%, P<0001 for all). Overall, 12,129 (27.7%) patients underwent AVR for ssAS. Females were less likely to undergo AVR compared with males (24.1% vs. 31.0%, adjusted hazard ratio [HR] 0.80, 95% confidence interval [CI] 0.77-0.83), but when treated, were more likely to undergo TAVR (37.9% vs. 30.9%, adjusted HR 1.21, 95% CI 1.15-1.27). Untreated females and males had similarly high rates of mortality at 1 year (31.1% vs. 31.3%, adjusted HR 0.98, 95% CI 0.94-1.03). Among those undergoing AVR, females had significantly higher mortality (10.2% vs. 9.4%, adjusted HR 1.24, 95% CI 1.10-1.41), driven by increased SAVR-associated mortality (9.0% vs. 7.6%, adjusted HR 1.43, 95% CI 1.21-1.69).

CONCLUSIONS

Treatment rates for ssAS patients remain suboptimal with disparities in female treatment.

Using Natural Language Processing to Measure and Improve Quality of Diabetes Care: A Systematic Review

BACKGROUND

Real-world evidence research plays an increasingly important role in diabetes care. However, a large fraction of real-world data are "locked" in narrative format. Natural language processing (NLP) technology offers a solution for analysis of narrative electronic data.

METHODS

We conducted a systematic review of studies of NLP technology focused on diabetes. Articles published prior to June 2020 were included.

RESULTS

We included 38 studies in the analysis. The majority (24; 63.2%) described only development of NLP tools; the remainder used NLP tools to conduct clinical research. A large fraction (17; 44.7%) of studies focused on identification of patients with diabetes; the rest covered a broad range of subjects that included hypoglycemia, lifestyle counseling, diabetic kidney disease, insulin therapy and others. The mean F₁ score for all studies where it was available was 0.882. It tended to be lower (0.817) in studies of more linguistically complex concepts. Seven studies reported findings with potential implications for improving delivery of diabetes care.

CONCLUSION

Research in NLP technology to study diabetes is growing quickly, although challenges (e.g. in analysis of more linguistically complex concepts) remain. Its potential to deliver evidence on treatment and improving quality of diabetes care is demonstrated by a number of studies. Further growth in this area would be aided by deeper collaboration between developers and end-users of natural language processing tools as well as by broader sharing of the tools themselves and related resources.

The Need for Ethnoracial Equity in Artificial Intelligence for Diabetes Management: Review and Recommendations

There is clear evidence to suggest that diabetes does not affect all populations equally. Among adults living with diabetes, those from ethnoracial minority communities—foreign-born, immigrant, refugee, and culturally marginalized—are at increased risk of poor health outcomes. Artificial intelligence (AI) is actively being researched as a means of improving diabetes management and care; however, several factors may predispose AI to ethnoracial bias. To better understand whether diabetes AI interventions are being designed in an ethnoracially equitable manner, we conducted a secondary analysis of 141 articles included in a 2018 review by Contreras and Vehi entitled “Artificial Intelligence for Diabetes Management and Decision Support: Literature Review.” Two members of our research team independently reviewed each article and selected those reporting ethnoracial data for further analysis. Only 10 articles (7.1%) were ultimately selected for secondary analysis in our case study. Of the 131 excluded articles, 118 (90.1%) failed to mention participants’ ethnic or racial backgrounds. The included articles reported ethnoracial data under various categories, including race (n=6), ethnicity (n=2), race/ethnicity (n=3), and percentage of Caucasian participants (n=1). Among articles specifically reporting race, the average distribution was 69.5% White, 17.1% Black, and 3.7% Asian. Only 2 articles reported inclusion of Native American participants. Given the clear ethnic and racial differences in diabetes biomarkers, prevalence, and outcomes, the inclusion of ethnoracial training data is likely to improve the accuracy of predictive models. Such considerations are imperative in AI-based tools, which are predisposed to negative biases due to their black-box nature and proneness to distributional shift. Based on our findings, we propose a short questionnaire to assess ethnoracial equity in research describing AI-based diabetes interventions. At this unprecedented time in history, AI can either mitigate or exacerbate disparities in health care. Future accounts of the infancy of diabetes AI must reflect our early and decisive action to confront ethnoracial inequities before they are coded into our systems and perpetuate the very biases we aim to eliminate. If we take deliberate and meaningful steps now toward training our algorithms to be ethnoracially inclusive, we can architect innovations in diabetes care that are bound by the diverse fabric of our society.

Incidence and prevalence of hypoglycaemia in type 1 and type 2 diabetes individuals: A systematic review and meta-analysis

BACKGROUND

Previous meta-analysis investigating the incidence and prevalence of hypoglycaemia in both types of diabetes is limited. The purpose of this review is to conduct a systematic review and meta-analysis of the existing literature which investigates the incidence and prevalence of hypoglycaemia in individuals with diabetes.

METHODS

PubMed, Embase and Cochrane library databases were searched up to October 2018. Observational studies including individuals with diabetes of all ages and reporting incidence and/or prevalence of hypoglycaemia were included. Two reviewers independently screened articles, extracted data and assessed the quality of included studies. Meta-analysis was performed using a random effects model with 95% confidence interval (CI) to estimate the pooled incidence and prevalence of hypoglycaemia in individuals with diabetes.

RESULTS

Our search strategy generated 35,007 articles, of which 72 studies matched the inclusion criteria and were included in the meta-analysis. The prevalence of hypoglycaemia ranged from 0.074% to 73.0%, comprising a total of 2,462,810 individuals with diabetes. The incidence rate of hypoglycaemia ranged from 0.072 to 42,890 episodes per 1,000 person-years: stratified by type of diabetes, it ranged from 14.5 to 42,890 episodes per 1,000 person-years and from 0.072 to 16,360 episodes per 1,000-person years in type 1 and type 2 diabetes, respectively.

CONCLUSION

Hypoglycaemia is very common among individuals with diabetes. Further studies are needed to investigate hypoglycaemia-associated risk factors.

Association of Pain on Hospital Discharge with the Risk of 30-Day Readmission in Patients with Total Hip and Knee Replacement

BACKGROUND

It is not clear if there is a risk of 30-day readmissions following total hip and knee arthroplasty in patients reporting high levels of pain at hospital discharge. We examined the relationship between post-surgical pain on the day of discharge and 30-day readmission in patients who received total knee and hip arthroplasty.

METHODS

Retrospective cohort study was conducted of patients who received total knee (n = 155,284) or hip arthroplasty (n = 89,283) from 2011 to 2018 using electronic health records from the Optum database. Four categories of pain at discharge were created, from none to severe. Multivariate logistic regression models to predict 30-day all-cause readmission were adjusted for patient and clinical characteristics and built separately for knee and hip arthroplasty patients.

RESULTS

Mean ages for hip and knee patients were 64.4 (standard deviation 11.3) and 65.7 (standard deviation 9.7) years, respectively. The majority of patients were female (hip: 54.4%; knee: 61.5%). The unadjusted rate of 30-day readmission was 3.54% for hip replacement and 3.66% for knee replacement. In models adjusted for patient and clinical characteristics, for patients with total hip replacement, the odds of 30-day readmission for those with severe pain score at discharge vs those with no pain at discharge were 1.60 (95% confidence interval 1.33-1.92). Similarly, readmission likelihood increased as pain at discharge increased (severe pain vs no pain) for patients with total knee arthroplasty (odds ratio 1.38, 95% confidence interval 1.19-1.59).

CONCLUSION

Our findings demonstrated that the pain scores on the day of discharge are associated with 30-day hospital readmission.

Association of Race/Ethnicity and Sex With Differences in Health Care Use and Treatment for Acne

Importance

Our understanding of potential racial/ethnic, sex, and other differences in health care use and treatment for acne is limited.

Objective

To identify potential disparities in acne care by evaluating factors associated with health care use and specific treatments for acne.

Design, Setting, and Participants

This retrospective cohort study used the Optum deidentified electronic health record data set to identify patients treated for acne from January 1, 2007, to June 30, 2017. Patients had at least 1 International Classification of Diseases, Ninth Revision (ICD-9) or International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code for acne and at least 1 year of continuous enrollment after the first diagnosis of acne. Data analysis was performed from September 1, 2019, to November 20, 2019.

Main Outcomes and Measures

Multivariable regression was used to quantify associations between basic patient demographic and socioeconomic characteristics and the outcomes of health care use and treatment for acne during 1 year of follow-up.

Results

A total of 29 928 patients (median [interquartile range] age, 20.2 [15.4-34.9] years; 19 127 [63.9%] female; 20 310 [67.9%] white) met the inclusion criteria for the study. Compared with non-Hispanic white patients, non-Hispanic black patients were more likely to be seen by a dermatologist (odds ratio [OR], 1.20; 95% CI, 1.09-1.31) but received fewer prescriptions for acne medications (incidence rate ratio, 0.89; 95% CI, 0.84-0.95). Of the acne treatment options, non-Hispanic black patients were more likely to receive prescriptions for topical retinoids (OR, 1.25; 95% CI, 1.14-1.38) and topical antibiotics (OR, 1.35; 95% CI, 1.21-1.52) and less likely to receive prescriptions for oral antibiotics (OR, 0.80; 95% CI, 0.72-0.87), spironolactone (OR, 0.68; 95% CI, 0.49-0.94), and isotretinoin (OR, 0.39; 95% CI, 0.23-0.65) than non-Hispanic white patients. Male patients were more likely to be prescribed isotretinoin than female patients (OR, 2.44; 95% CI, 2.01-2.95). Compared with patients with commercial insurance, those with Medicaid were less likely to see a dermatologist (OR, 0.46; 95% CI, 0.41-0.52) or to be prescribed topical retinoids (OR, 0.82; 95% CI, 0.73-0.92), oral antibiotics (OR, 0.87; 95% CI, 0.79-0.97), spironolactone (OR, 0.50; 95% CI, 0.31-0.80), and isotretinoin (OR, 0.43; 95% CI, 0.25-0.75).

Conclusions and Relevance

The findings identify racial/ethnic, sex, and insurance-based differences in health care use and prescribing patterns for acne that are independent of other sociodemographic factors and suggest potential disparities in acne care. In particular, the study found underuse of systemic therapies among racial/ethnic minorities and isotretinoin among female patients with acne. Further study is needed to confirm and understand the reasons for these differences.

Age-appropriate cancer screening: A cohort study of adults with psoriasis prescribed biologics, adults in the general population, and adults with hypertension

BACKGROUND

Psoriasis is associated with increased risk of developing and dying from cancer.

OBJECTIVE

To evaluate whether psoriasis patients who are prescribed biologics receive the recommended screening for cervical, breast, and colon cancer.

METHODS

We conducted a retrospective cohort study using the Optum deidentified Electronic Health Record data set. Incidence rates for cervical, breast, and colon cancer screening were compared between psoriasis patients who were prescribed biologics and 2 matched comparator cohorts: general patient population and patients being managed for hypertension. Multivariable Cox proportional hazards regression was performed to assess for differences in the rates of cancer screening.

RESULTS

Compared with those in the general population without psoriasis, psoriasis patients who were prescribed biologics had higher screening rates for cervical cancer (adjusted hazard ratio [aHR] 1.09; 95% confidence interval [CI] 1.02-1.16) and colon cancer (aHR 1.10; 95% CI 1.02-1.18). Compared with those with hypertension, patients with psoriasis who were prescribed biologics had lower screening rates for breast cancer (aHR 0.88; 95% CI 0.83-0.94) and colon cancer (aHR 0.89; 95% CI 0.83-0.95).

CONCLUSIONS AND RELEVANCE

Patients with psoriasis who are prescribed biologic therapies may not be receiving adequate age-appropriate cancer screening, especially for breast and colon cancer.

Racial Differences in the Use of Aortic Valve Replacement for Treatment of Symptomatic Severe Aortic Valve Stenosis in the Transcatheter Aortic Valve Replacement Era

Background Aortic valve replacement (AVR) is a life-saving treatment for patients with symptomatic severe aortic valve stenosis. We sought to determine whether transcatheter AVR has resulted in a more equitable treatment rate by race in the United States. Methods and Results A total of 32 853 patients with symptomatic severe aortic valve stenosis were retrospectively identified via Optum's deidentified electronic health records database (2007-2017). AVR rates in non-Hispanic Black and White patients were assessed in the year after diagnosis. Multivariate Fine-Gray hazards models were used to evaluate the likelihood of AVR by race, with adjustment for patient factors and the managing cardiologist. Time-trend and 1-year symptomatic severe aortic valve stenosis survival analyses were also performed. From 2011 to 2016, the rate of AVR increased from 20.1% to 37.1%. Overall, Black individuals were less likely than Whites to receive AVR (22.9% versus 31.0%; unadjusted hazard ratio [HR], 0.70; 95% CI, 0.62-0.79; fully adjusted HR, 0.76; 95% CI, 0.67-0.85). Yet, during 2015 to 2016, AVR racial differences were attenuated (29.5% versus 35.2%; adjusted HR, 0.86; 95% CI, 0.74-1.02) because of greater uptake of transcatheter AVR in Blacks than Whites (53.4% of AVRs versus 47.3%; P=0.128). Untreated patients had significantly higher 1-year mortality than those treated (adjusted HR, 0.57; 95% CI, 0.53-0.61), which was consistent by race (interaction P value=0.52). Conclusions Although transcatheter AVR has increased the use of AVR in the United States, treatment rates remain low. Black patients with symptomatic severe aortic valve stenosis were less likely than White patients to receive AVR, yet these differences have recently narrowed.

Natural Language Processing Improves Detection of Nonsevere Hypoglycemia in Medical Records Versus Coding Alone in Patients With Type 2 Diabetes but Does Not Improve Prediction of Severe Hypoglycemia Events: An Analysis Using the Electronic Medical Record in a Large Health System

OBJECTIVE

To determine if natural language processing (NLP) improves detection of nonsevere hypoglycemia (NSH) in patients with type 2 diabetes and no NSH documentation by diagnosis codes and to measure if NLP detection improves the prediction of future severe hypoglycemia (SH).

RESEARCH DESIGN AND METHODS

From 2005 to 2017, we identified NSH events by diagnosis codes and NLP. We then built an SH prediction model.

RESULTS

There were 204,517 patients with type 2 diabetes and no diagnosis codes for NSH. Evidence of NSH was found in 7,035 (3.4%) of patients using NLP. We reviewed 1,200 of the NLP-detected NSH notes and confirmed 93% to have NSH. The SH prediction model (C-statistic 0.806) showed increased risk with NSH (hazard ratio 4.44; P < 0.001). However, the model with NLP did not improve SH prediction compared with diagnosis code-only NSH.

CONCLUSIONS

Detection of NSH improved with NLP in patients with type 2 diabetes without improving SH prediction.

Evaluation of provider documentation patterns as a tool to deliver ongoing patient-centred diabetes education and support

BACKGROUND

Diabetes mellitus (DM) is one of the most common chronic diseases in the world. As a disease with long-term complications requiring changes in management, DM requires not only education at the time of diagnosis, but ongoing diabetes self-management education and support (DSME/S). In the United States, however, only a small proportion of people with DM receive DSME/S, although evidence supports benefits of ongoing DSME/S. The diabetes education that providers deliver during follow-up visits may be an important source for DSME/S for many people with DM.

METHODS

We collected 200 clinic notes of follow-up visits for 100 adults with DM and studied the History of Present Illness (HPI) and Impression and Plan (I&P) sections. Using a codebook based on the seven principles of American Association of Diabetes Educators Self-Care Behaviors (AADE7), we conducted a multi-step deductive thematic analysis to determine the patterns of DSME/S information occurrence in clinic notes. Additionally, we used the generalised linear mixed models for investigating whether providers delivered DSME/S to people with DM based on patient characteristics.

RESULTS

During follow-up visits, Monitoring was the most common self-care behaviour mentioned in both HPI and I&P sections. Being Active was the least common self-care behaviour mentioned in the HPI section and Healthy Coping was the least common self-care behaviour mentioned in the I&P section. We found providers delivered more information on Healthy Eating to men compared to women in I&P section. Generally, providers delivered DSME/S to people with DM regardless of patient characteristics.

CONCLUSIONS

This study focused on the frequency distribution of information providers delivered to the people with DM during follow-up clinic visits based on the AADE7. The results may indicate a lack of patient-centred education when people with DM visit providers for ongoing management. Further studies are needed to identify the underlying reasons why providers have difficulty delivering patient-centred education.

Documentation of hypoglycemia assessment among adults with diabetes during clinical encounters in primary care and endocrinology practices

PURPOSE

To examine the proportion of diabetes-focused clinical encounters in primary care and endocrinology practices where the evaluation for hypoglycemia is documented; and when it is, identify clinicians' stated actions in response to patient-reported events.

METHODS

A total of 470 diabetes-focused encounters among 283 patients nonpregnant adults (≥18 years) with type 1 or type 2 diabetes mellitus in this retrospective cohort study. Participants were randomly identified in blocks of treatment strategy and care location (95 and 52 primary care encounters among hypoglycemia-prone medications (i.e. insulin, sulfonylurea) and others patients, respectively; 94 and 42 endocrinology encounters among hypo-treated and others, respectively). Documentation of hypoglycemia and subsequent management plan in the electronic health record were evaluated.

RESULTS

Overall, 132 (46.6%) patients had documentation of hypoglycemia assessment, significantly more prevalent among hypo-treated patients seen in endocrinology than in primary care (72.3% vs. 47.4%; P = 0.001). Hypoglycemia was identified by patient in 38.2% of encounters. Odds of hypoglycemia assessment documentation was highest among the hypo-treated (OR 13.6; 95% CI 5.5-33.74, vs. others) and patients seen in endocrine clinic (OR 4.48; 95% CI 2.3-8.6, vs. primary care). After documentation of hypoglycemia, treatment was modified in 30% primary care and 46% endocrine clinic encounters; P = 0.31. Few patients were referred to diabetes self-management education and support (DSMES).

CONCLUSIONS

Continued efforts to improve hypoglycemia evaluation, documentation, and management are needed, particularly in primary care. This includes not only screening at-risk patients for hypoglycemia, but also modifying their treatment regimens and/or leveraging DSMES.

Assessing stroke severity using electronic health record data: a machine learning approach

Background

Stroke severity is an important predictor of patient outcomes and is commonly measured with the National Institutes of Health Stroke Scale (NIHSS) scores. Because these scores are often recorded as free text in physician reports, structured real-world evidence databases seldom include the severity. The aim of this study was to use machine learning models to impute NIHSS scores for all patients with newly diagnosed stroke from multi-institution electronic health record (EHR) data.

Methods

NIHSS scores available in the Optum© de-identified Integrated Claims-Clinical dataset were extracted from physician notes by applying natural language processing (NLP) methods. The cohort analyzed in the study consists of the 7149 patients with an inpatient or emergency room diagnosis of ischemic stroke, hemorrhagic stroke, or transient ischemic attack and a corresponding NLP-extracted NIHSS score. A subset of these patients (n = 1033, 14%) were held out for independent validation of model performance and the remaining patients (n = 6116, 86%) were used for training the model. Several machine learning models were evaluated, and parameters optimized using cross-validation on the training set. The model with optimal performance, a random forest model, was ultimately evaluated on the holdout set.

Results

Leveraging machine learning we identified the main factors in electronic health record data for assessing stroke severity, including death within the same month as stroke occurrence, length of hospital stay following stroke occurrence, aphagia/dysphagia diagnosis, hemiplegia diagnosis, and whether a patient was discharged to home or self-care. Comparing the imputed NIHSS scores to the NLP-extracted NIHSS scores on the holdout data set yielded an R² (coefficient of determination) of 0.57, an R (Pearson correlation coefficient) of 0.76, and a root-mean-squared error of 4.5.

Conclusions

Machine learning models built on EHR data can be used to determine proxies for stroke severity. This enables severity to be incorporated in studies of stroke patient outcomes using administrative and EHR databases.

Assessing Prevalence of Hypoglycemia in a Medical Transcription Database

PURPOSE

The prevalence of hypoglycemia in patients with diabetes mellitus is likely underreported, particularly with regard to non-severe episodes, and representative estimates require more detailed data than claims or typical electronic health record (EHR) databases provide. This study examines the prevalence of hypoglycemia as identified in a medical transcription database.

PATIENTS AND METHODS

The Amplity Insights database contains medical content dictated by providers detailing patient encounters with health care professionals (HCPs) from across the United States. Natural language processing (NLP) was used to identify episodes of hypoglycemia using both symptom-based and non-symptom-based definitions of hypoglycemic events. This study examined records of 41,688 patients with type 1 diabetes mellitus and 317,399 patients with type 2 diabetes mellitus between January 1, 2016, and April 30, 2018.

RESULTS

Using a non-symptom-based definition, the prevalence of hypoglycemia was 18% among patients with T1DM and 8% among patients with T2DM. These estimates show the prevalence of hypoglycemia to be 2- to 9-fold higher than the 1% to 4% prevalence estimates suggested by claims database analyses.

CONCLUSION

In this exploration of a medical transcription database, the prevalence of hypoglycemia was considerably higher than what has been reported via retrospective analyses from claims and EHR databases. This analysis suggests that data sources other than claims and EHR may provide a more in-depth look into discrepancies between the mention of hypoglycemia events during a health care visit and documentation of hypoglycemia in patient records.

Real-world Effectiveness of Liraglutide vs. Sitagliptin Among Older Patients with Type 2 Diabetes Enrolled in a Medicare Advantage Prescription Drug Plan: A Retrospective Observational Study

INTRODUCTION

Liraglutide and sitagliptin were compared on glycemic control and all-cause healthcare costs over a 1-year period among older adults with type 2 diabetes (65-89 years) enrolled in a national Medicare Advantage Prescription Drug health plan.

METHODS

This was a retrospective study in which the index date was the first prescription fill for liraglutide or sitagliptin between 25 January 2010 and 31 December 2014. Post-index treatment persistence and glycosylated hemoglobin (HbA1_c) at baseline and 1 year (± 90 days) post-index date were required. Patients were excluded if their record included use of insulin during the baseline period. Inverse probability of treatment weighting using stabilized weights was employed with final covariate adjusted regression modeling to estimate the primary outcome (mean change in HbA1_c) and secondary outcomes (achieving glycemic goal and costs), each at 1-year post-index date.

RESULTS

Overall, 3056 patients met the selection criteria, of whom 218 filled prescriptions for liraglutide and 2838 for sitagliptin. Adjusted mean change in HbA1_c at 1 year post-index was - 0.42 with liraglutide versus - 0.12 with sitagliptin (P  = 0.0012). Adjusted odds of achieving the treatment goals of HbA1_c < 7% and achieving an HbA1_c reduction of ≥ 1% were higher for  those on liraglutide than for those on sitagliptin (1.68, 95% confidence interval [CI] 1.25-2.24 and 1.76, 95% CI 1.31-2.36), respectively. Total healthcare costs in those achieving an HbA1_c of < 7% were not significantly different between treatment groups but were higher within the liraglutide group for those achieving an HbA1_c < 8%.

CONCLUSIONS

When compared to sitagliptin, liraglutide was associated with greater achievement of an HbA1_c < 7% over a 1-year period in an older population. This finding was not associated with a statistically significant increase in all-cause total healthcare costs, although costs were slightly higher in the liraglutide group than in the sitagliptin group.

A systematic review of natural language processing for classification tasks in the field of incident reporting and adverse event analysis

CONTEXT

Adverse events in healthcare are often collated in incident reports which contain unstructured free text. Learning from these events may improve patient safety. Natural language processing (NLP) uses computational techniques to interrogate free text, reducing the human workload associated with its analysis. There is growing interest in applying NLP to patient safety, but the evidence in the field has not been summarised and evaluated to date.

OBJECTIVE

To perform a systematic literature review and narrative synthesis to describe and evaluate NLP methods for classification of incident reports and adverse events in healthcare.

METHODS

Data sources included Medline, Embase, The Cochrane Library, CINAHL, MIDIRS, ISI Web of Science, SciELO, Google Scholar, PROSPERO, hand searching of key articles, and OpenGrey. Data items were manually abstracted to a standardised extraction form.

RESULTS

From 428 articles screened for eligibility, 35 met the inclusion criteria of using NLP to perform a classification task on incident reports, or with the aim of detecting adverse events. The majority of studies used free text from incident reporting systems or electronic health records. Models were typically designed to classify by type of incident, type of medication error, or harm severity. A broad range of NLP techniques are demonstrated to perform these classification tasks with favourable performance outcomes. There are methodological challenges in how these results can be interpreted in a broader context.

CONCLUSION

NLP can generate meaningful information from unstructured data in the specific domain of the classification of incident reports and adverse events. Understanding what or why incidents are occurring is important in adverse event analysis. If NLP enables these insights to be drawn from larger datasets it may improve the learning from adverse events in healthcare.

Text preprocessing for improving hypoglycemia detection from clinical notes - A case study of patients with diabetes

BACKGROUND AND OBJECTIVE

Hypoglycemia is a common safety event when attempting to optimize glycemic control in diabetes (DM). While electronic medical records provide a natural ground for detecting and analyzing hypoglycemia, ICD codes used in the databases may be invalid, insensitive or non-specific in detecting new hypoglycemic events. We developed text preprocessing methods to improve automatic detection of hypoglycemia from analysis of clinical encounter text notes.

METHODS

We set out to improve hypoglycemia detection from clinical notes by introducing three preprocessing methods: stop word filtering, medication signaling, and ICD narrative enrichment. To test the proposed methods, we selected clinical notes from VA Maryland Healthcare System, based on various combinations of three criteria that are suggestive of hypoglycemia, including ICD-9 code of diabetes and hypoglycemia, laboratory glucose values < 70 md/dL, and text reference to a proximate hypoglycemia event. In addition, we constructed one dataset of 395 clinical notes from year 2009 and another of 460 notes from year 2014 to test the generality of the proposed methods. For each of the datasets, two physician judges manually reviewed individual clinical notes to determine whether hypoglycemia was present or absent. A third physician judge served as a final adjudicator for disagreements.

RESULTS

Each of the proposed preprocessing methods contributed to the performance of hypoglycemia detection by significantly increasing the F1 score in the range of 5.3∼7.4% on one dataset (p < .01). Among the methods, stop word filtering contributed most to the performance improvement (7.4%). Combining all the preprocessing methods led to greater performance gain (p < .001) compared with using each method individually. Similar patterns were observed for the other dataset with the F1 score being increased in the range of 7.7%∼9.4% by individual methods (p < .001). Nevertheless, combining the three methods did not yield additional performance gain.

CONCLUSION

The proposed text preprocessing methods improved the performance of hypoglycemia detection from clinical text notes. Stop word filtering achieved the most performance improvement. ICD narrative enrichment boosted the recall of detection. Combining the three preprocessing methods led to additional performance gains.

Reporting of hypoglycaemia in clinical trials of basal insulins: A need for consensus

Hypoglycaemia is a common side-effect of diabetes therapies, particularly insulin, and imposes a substantial burden on individuals and healthcare systems. Consequently, regulatory approval of newer basal insulin (BI) therapies has relied on demonstration of a balance between achievement of good glycaemic control and less hypoglycaemia. Randomized controlled trials (RCTs) are the gold standard for assessing efficacy and safety, including hypoglycaemia risk, of BIs and are invaluable for obtaining regulatory approval. However, their highly selected patient populations and their conditions lead to results that may not be representative of real-life situations. Real-world evidence (RWE) studies are more representative of clinical practice, but they also have limitations. As such, data both from RCTs and RWE studies provide a fuller picture of the hypoglycaemia risk with BI therapies. However, substantial differences exist in the way hypoglycaemia is reported across these studies, which confounds comparisons of hypoglycaemia frequency among different BIs. This problem is ongoing and persists in recent trials of second-generation BI analogues. Although they provide a lower risk of hypoglycaemia when compared with earlier BIs, they do not eliminate it. This review describes differences in the way hypoglycaemia is reported across RCTs and RWE studies of second-generation BI analogues and examines potential reasons for these differences. For studies of BIs, there is a need to standardize aspects of design, analysis and methods of reporting to better enable interpretation of the efficacy and safety of such insulins among studies; such aspects include length of follow-up, glycaemic targets, hypoglycaemia definitions and time intervals for determining nocturnal events.

The clinical utility of laboratory monitoring during isotretinoin therapy for acne and changes to monitoring practices over time

BACKGROUND

As a result of concerns about hypertriglyceridemia, liver enzyme abnormalities, and leukopenia during isotretinoin therapy for acne, patients are often monitored closely with routine laboratory assessments, although the value of this practice has been questioned.

METHODS

We conducted a cohort study of patients receiving isotretinoin for acne between January 1, 2008, and June 30, 2017, using the OptumInsights Electronic Health Record Database (Optum, Eden Prairie, MN) to evaluate the frequency of laboratory abnormalities. Poisson regression was used to evaluate for changes to the frequency of routine laboratory monitoring over time.

RESULTS

Among 1863 patients treated with isotretinoin, grade 3 or greater triglyceride and liver function testing abnormalities were noted in fewer than 1% and 0.5% of patients screened, respectively. No grade 3 or greater cholesterol or complete blood count abnormalities were observed. There were no meaningful changes in the frequency of laboratory monitoring over time.

LIMITATIONS

Limitations include that we are unable to evaluate the clinical notes to understand the exact clinical decision making when clinicians encountered abnormal laboratory values.

CONCLUSION

Although laboratory abnormalities are rare and often do not influence management, frequent laboratory monitoring remains a common practice. There are opportunities to improve the quality of care among patients being treated with isotretinoin for acne by reducing the frequency of lipid and liver function monitoring and by eliminating complete blood count monitoring.

Natural Language Processing of Clinical Notes on Chronic Diseases: Systematic Review

Background

Novel approaches that complement and go beyond evidence-based medicine are required in the domain of chronic diseases, given the growing incidence of such conditions on the worldwide population. A promising avenue is the secondary use of electronic health records (EHRs), where patient data are analyzed to conduct clinical and translational research. Methods based on machine learning to process EHRs are resulting in improved understanding of patient clinical trajectories and chronic disease risk prediction, creating a unique opportunity to derive previously unknown clinical insights. However, a wealth of clinical histories remains locked behind clinical narratives in free-form text. Consequently, unlocking the full potential of EHR data is contingent on the development of natural language processing (NLP) methods to automatically transform clinical text into structured clinical data that can guide clinical decisions and potentially delay or prevent disease onset.

Objective

The goal of the research was to provide a comprehensive overview of the development and uptake of NLP methods applied to free-text clinical notes related to chronic diseases, including the investigation of challenges faced by NLP methodologies in understanding clinical narratives.

Methods

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were followed and searches were conducted in 5 databases using “clinical notes,” “natural language processing,” and “chronic disease” and their variations as keywords to maximize coverage of the articles.

Results

Of the 2652 articles considered, 106 met the inclusion criteria. Review of the included papers resulted in identification of 43 chronic diseases, which were then further classified into 10 disease categories using the International Classification of Diseases, 10th Revision. The majority of studies focused on diseases of the circulatory system (n=38) while endocrine and metabolic diseases were fewest (n=14). This was due to the structure of clinical records related to metabolic diseases, which typically contain much more structured data, compared with medical records for diseases of the circulatory system, which focus more on unstructured data and consequently have seen a stronger focus of NLP. The review has shown that there is a significant increase in the use of machine learning methods compared to rule-based approaches; however, deep learning methods remain emergent (n=3). Consequently, the majority of works focus on classification of disease phenotype with only a handful of papers addressing extraction of comorbidities from the free text or integration of clinical notes with structured data. There is a notable use of relatively simple methods, such as shallow classifiers (or combination with rule-based methods), due to the interpretability of predictions, which still represents a significant issue for more complex methods. Finally, scarcity of publicly available data may also have contributed to insufficient development of more advanced methods, such as extraction of word embeddings from clinical notes.

Conclusions

Efforts are still required to improve (1) progression of clinical NLP methods from extraction toward understanding; (2) recognition of relations among entities rather than entities in isolation; (3) temporal extraction to understand past, current, and future clinical events; (4) exploitation of alternative sources of clinical knowledge; and (5) availability of large-scale, de-identified clinical corpora.

Rates of Hypoglycemia Predicted in Patients with Type 2 Diabetes on Insulin Glargine 300 U/ml Versus First- and Second-Generation Basal Insulin Analogs: The Real-World LIGHTNING Study

INTRODUCTION

The LIGHTNING study applied conventional and advanced analytic approaches to model, predict, and compare hypoglycemia rates of people with type 2 diabetes (T2DM) on insulin glargine 300 U/ml (Gla-300) with those on first-generation (insulin glargine 100 U/ml [Gla-100]; insulin detemir [IDet]) or second-generation (insulin degludec [IDeg]) basal-insulin (BI) analogs, utilizing a large real-world database.

METHODS

Data were collected between 1 January 2007 and 31 March 2017 from the Optum Humedica US electronic health records [EHR] database. Patient-treatments, the period during which a patient used a specific BI, were analyzed for patients who switched from a prior BI or those who newly initiated BI therapy. Data were analyzed using two approaches: propensity score matching (PSM) and a predictive modeling approach using machine learning.

RESULTS

A total of 831,456 patients with T2DM receiving BI were included from the EHR data set. Following selection, 198,198 patient-treatments were available for predictive modeling. The analysis showed that rates of severe hypoglycemia (using a modified definition) were approximately 50% lower with Gla-300 than with Gla-100 or IDet in insulin-naïve individuals, and 30% lower versus IDet in BI switchers (all p < 0.05). Similar rates of severe hypoglycemia were predicted for Gla-300 and IDeg, regardless of prior insulin experience. Similar results to those observed in the overall cohorts were seen in analyses across subgroups at a particularly high risk of hypoglycemia. PSM (performed on 157,573 patient-treatments) revealed comparable reductions in HbA_1c with Gla-300 versus first- and second-generation BI analogs, alongside lower rates of severe hypoglycemia with Gla-300 versus first-generation BI analogs (p < 0.05) and similar rates versus IDeg in insulin-naïve and BI-switcher cohorts.

CONCLUSIONS

Based on real-world data, predicted rates of severe hypoglycemia with Gla-300 tended to be lower versus first-generation BI analogs and similar versus IDeg in a wide spectrum of patients with T2DM.

FUNDING

Sanofi, Paris, France.

Predictive Modeling of Hypoglycemia Risk with Basal Insulin Use in Type 2 Diabetes: Use of Machine Learning in the LIGHTNING Study

INTRODUCTION

Hypoglycemia remains a global burden and a limiting factor in the glycemic management of people with diabetes using basal insulins or oral antihyperglycemic drugs. Hypoglycemia data gleaned from randomized controlled trials (RCTs) have limited generalizability, as the strict RCT methodology and inclusion criteria do not fully reflect the real-world clinical picture. Therefore, real-world evidence, gathered from sources including electronic health records (EHR), is increasingly recognized as an important adjunct to RCTs.

AIMS AND METHODS

The LIGHTNING study applied advanced analytical methods, including machine learning (ML), to EHR data. The study aimed to predict hypoglycemic event rates in patients with type 2 diabetes (T2DM) receiving different basal insulin treatments to identify potential subgroups of patients who are at lower risk of hypoglycemia when treated with one basal insulin compared with another and to predict hypoglycemia-related cost savings in these subgroups. Here we provide an overview of the objectives, study design and methods, and validation approaches used in the LIGHTNING study.

CONCLUSION

It is hoped that results of the LIGHTNING study will help facilitate real-world clinical decision-making in addition to providing a clinically relevant predictive model of hypoglycemia risk.

FUNDING

Sanofi.

U.S. prevalence of endocrine therapy-naïve locally advanced or metastatic breast cancer

Background

Variations in treatment choice, or late stage at first diagnosis, mean that, despite guideline recommendations, not all patients with hormone receptor (hr)-positive locally advanced or metastatic breast cancer (la/mbca) will have received endocrine therapy before disease progression. In the present study, we aimed to estimate the proportion of women with postmenopausal hr-positive la/mbca in the United States who are endocrine therapy-naïve.

Methods

Women in the Optum Electronic Health Record (ehr) database with a breast cancer (bca) diagnosis (January 2008-March 2015) were included. Patient and malignancy characteristics were identified using structured data fields and natural-language processing of free-text clinical notes. The proportion of women with postmenopausal hr-positive, human epidermal growth factor 2 (her2)-negative (or unknown) la/mbca who had not received prior endocrine therapy was determined. Results were extrapolated to the entire U.S. population using the U.S. National Cancer Institute's Surveillance, Epidemiology, and End Results database. Results are presented descriptively.

Results

In the ehr database, 11,831 women with bca had discernible information on postmenopausal status, hr status, and disease stage. Of those women, 1923 (16.3%) had postmenopausal hr-positive, her2-negative (or unknown) la/mbca, and 70.7% of those 1923 patients (n = 1360) had not received prior endocrine therapy, accounting for 11.5% of the overall population. Extrapolating those estimates nationally suggests an annual incidence of 14,784 cases, and a 5-year limited duration prevalence of 50,638 cases.

Conclusions

A substantial proportion of women with postmenopausal hr-positive la/mbca in the United States could be endocrine therapy-naïve.

Association of metformin, sulfonylurea and insulin use with brain structure and function and risk of dementia and Alzheimer's disease: Pooled analysis from 5 cohorts

Objective

To determine whether classes of diabetes medications are associated with cognitive health and dementia risk, above and beyond their glycemic control properties.

Research design and methods

Findings were pooled from 5 population-based cohorts: the Framingham Heart Study, the Rotterdam Study, the Atherosclerosis Risk in Communities (ARIC) Study, the Aging Gene-Environment Susceptibility-Reykjavik Study (AGES) and the Sacramento Area Latino Study on Aging (SALSA). Differences between users and non-users of insulin, metformin and sulfonylurea were assessed in each cohort for cognitive and brain MRI measures using linear regression models, and cognitive decline and dementia/AD risk using mixed effect models and Cox regression analyses, respectively. Findings were then pooled using meta-analytic techniques, including 3,590 individuals with diabetes for the prospective analysis.

Results

After adjusting for potential confounders including indices of glycemic control, insulin use was associated with increased risk of new-onset dementia (pooled HR (95% CI) = 1.58 (1.18, 2.12);p = 0.002) and with a greater decline in global cognitive function (β = -0.014±0.007;p = 0.045). The associations with incident dementia remained similar after further adjustment for renal function and excluding persons with diabetes whose treatment was life-style change only. Insulin use was not related to cognitive function nor to brain MRI measures. No significant associations were found between metformin or sulfonylurea use and outcomes of brain function and structure. There was no evidence of significant between-study heterogeneity.

Conclusions

Despite its advantages in controlling glycemic dysregulation and preventing complications, insulin treatment may be associated with increased adverse cognitive outcomes possibly due to a greater risk of hypoglycemia.

Natural Language Processing for EHR-Based Computational Phenotyping

This article reviews recent advances in applying natural language processing (NLP) to Electronic Health Records (EHRs) for computational phenotyping. NLP-based computational phenotyping has numerous applications including diagnosis categorization, novel phenotype discovery, clinical trial screening, pharmacogenomics, drug-drug interaction (DDI), and adverse drug event (ADE) detection, as well as genome-wide and phenome-wide association studies. Significant progress has been made in algorithm development and resource construction for computational phenotyping. Among the surveyed methods, well-designed keyword search and rule-based systems often achieve good performance. However, the construction of keyword and rule lists requires significant manual effort, which is difficult to scale. Supervised machine learning models have been favored because they are capable of acquiring both classification patterns and structures from data. Recently, deep learning and unsupervised learning have received growing attention, with the former favored for its performance and the latter for its ability to find novel phenotypes. Integrating heterogeneous data sources have become increasingly important and have shown promise in improving model performance. Often, better performance is achieved by combining multiple modalities of information. Despite these many advances, challenges and opportunities remain for NLP-based computational phenotyping, including better model interpretability and generalizability, and proper characterization of feature relations in clinical narratives.

Patient Characteristics Associated With Severe Hypoglycemia in a Type 2 Diabetes Cohort in a Large, Integrated Health Care System From 2006 to 2015

OBJECTIVE

To identify severe hypoglycemia events, defined as emergency department visits or hospitalizations for hypoglycemia, in patients with type 2 diabetes receiving care in a large health system and to identify patient characteristics associated with severe hypoglycemia events.

RESEARCH DESIGN AND METHODS

This was a retrospective cohort study from January 2006 to December 2015 using the electronic medical record in the Cleveland Clinic Health System (CCHS). Participants included 50,439 patients with type 2 diabetes receiving care in the CCHS. Number of severe hypoglycemia events and associated patient characteristics were identified.

RESULTS

The incidence proportion of severe hypoglycemia increased from 0.12% in 2006 to 0.31% in 2015 (P = 0.01). Compared with patients who did not experience severe hypoglycemia, those with severe hypoglycemia had similar median glycosylated hemoglobin (HbA_1c) levels. More patients with severe hypoglycemia versus those without had a prior diagnosis of nonsevere hypoglycemia (9% vs. 2%, P < 0.001). Logistic regression confirmed an increased odds for severe hypoglycemia with insulin, sulfonylureas, increased number of diabetes medications, history of nonsevere hypoglycemia (odds ratio [OR] 3.01, P < 0.001), HbA_1c <6% (42 mmol/mol) (OR 1.95, P < 0.001), black race, and increased Charlson comorbidity index. Lower odds of severe hypoglycemia were noted with higher BMI and use of metformin, dipeptidyl peptidase 4 inhibitors, and glucagon-like peptide 1 agonists.

CONCLUSIONS

In this retrospective study of patients with type 2 diabetes with severe hypoglycemia, patient characteristics were identified. Patients with severe hypoglycemia had previous nonsevere hypoglycemia diagnoses more frequently than those without. Identifying patients at high risk at the point of care can allow for change in modifiable risk factors and prevention of severe hypoglycemia events.

Artificial Intelligence for Diabetes Management and Decision Support: Literature Review

Background

Artificial intelligence methods in combination with the latest technologies, including medical devices, mobile computing, and sensor technologies, have the potential to enable the creation and delivery of better management services to deal with chronic diseases. One of the most lethal and prevalent chronic diseases is diabetes mellitus, which is characterized by dysfunction of glucose homeostasis.

Objective

The objective of this paper is to review recent efforts to use artificial intelligence techniques to assist in the management of diabetes, along with the associated challenges.

Methods

A review of the literature was conducted using PubMed and related bibliographic resources. Analyses of the literature from 2010 to 2018 yielded 1849 pertinent articles, of which we selected 141 for detailed review.

Results

We propose a functional taxonomy for diabetes management and artificial intelligence. Additionally, a detailed analysis of each subject category was performed using related key outcomes. This approach revealed that the experiments and studies reviewed yielded encouraging results.

Conclusions

We obtained evidence of an acceleration of research activity aimed at developing artificial intelligence-powered tools for prediction and prevention of complications associated with diabetes. Our results indicate that artificial intelligence methods are being progressively established as suitable for use in clinical daily practice, as well as for the self-management of diabetes. Consequently, these methods provide powerful tools for improving patients’ quality of life.

Hypoglycaemia seriousness and weight gain as determinants of cardiovascular disease outcomes among sulfonylurea users

AIMS

Certain treatments for type 2 diabetes mellitus cause hypoglycaemia and weight gain, and thus might counteract the benefits of intensive glucose control. We quantify the association of cardiovascular disease (CVD) outcomes with hypoglycaemia and weight gain among patients with type 2 diabetes treated with sulfonylureas.

MATERIALS AND METHODS

This cohort study included patients from January 2009 through December 2014 who were selected from within a deidentified nationwide electronic health records repository, including multiple provider networks and electronic medical records systems. Hypoglycaemia measures from structured data fields and free text clinical notes were categorized as serious or non-serious. Covariate adjusted Poisson regression analysis was used to assess the association between frequency of hypoglycaemia (by severity), or magnitude of weight change, and incidence of acute myocardial infarction (AMI), congestive heart failure (CHF) and stroke.

RESULTS

Among 143 635 eligible patients, we observed 5669 cases of AMI, 14 109 incident cases of CHF and 7017 cases of stroke. Overall incidence rates were 1.53, 4.26 and 1.92 per 100 person-years for AMI, CHF and stroke, respectively. The associations between overall hypoglycaemia and each of the CVD outcomes were positive, with stronger associations observed for serious hypoglycaemia and attenuated or null associations observed for non-serious hypoglycaemia. Weight change exhibited a U-shaped association with increased risks associated with both weight loss and weight gain relative to stable weight.

CONCLUSIONS

This study provides evidence of increased CVD risk associated with hypoglycaemia, especially with serious hypoglycaemia events. While associations were attenuated with non-serious hypoglycaemia, the results were suggestive of a potential increased risk.