Machine learning to identify chronic cough from administrative claims data

Accurate identification of patient populations is an essential component of clinical research, especially for medical conditions such as chronic cough that are inconsistently defined and diagnosed. We aimed to develop and compare machine learning models to identify chronic cough from medical and pharmacy claims data. In this retrospective observational study, we compared 3 machine learning algorithms based on XG Boost, logistic regression, and neural network approaches using a large claims and electronic health record database. Of the 327,423 patients who met the study criteria, 4,818 had chronic cough based on linked claims-electronic health record data. The XG Boost model showed the best performance, achieving a Receiver-Operator Characteristic Area Under the Curve (ROC-AUC) of 0.916. We selected a cutoff that favors a high positive predictive value (PPV) to minimize false positives, resulting in a sensitivity, specificity, PPV, and negative predictive value of 18.0%, 99.6%, 38.7%, and 98.8%, respectively on the held-out testing set (n = 82,262). Logistic regression and neural network models achieved slightly lower ROC-AUCs of 0.907 and 0.838, respectively. The XG Boost and logistic regression models maintained their robust performance in subgroups of individuals with higher rates of chronic cough. Machine learning algorithms are one way of identifying conditions that are not coded in medical records, and can help identify individuals with chronic cough from claims data with a high degree of classification value.

Systematic literature review of treatments used for refractory or unexplained chronic cough in adults

BACKGROUND

Refractory or unexplained chronic cough (RCC or UCC) is difficult to manage and is usually treated by the off-label use of drugs approved for other indications.

OBJECTIVE

The objectives of this systematic literature review (SLR) were to identify and characterize the current published body of evidence for the efficacy and safety of treatments for RCC or UCC.

METHODS

The SLR was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The SLRs pre-defined population included patients ≥18 years of age who were diagnosed with chronic cough. The review was not restricted to any intervention type or study comparator, nor by timeframe.

RESULTS

A total of 20 eligible publications from 19 unique trials were included. Seventeen of these trials were randomized controlled trials and most (14/17) were placebo-controlled. There was considerable variability between trials in the definition of RCC or UCC, participant exclusion and inclusion criteria, outcome measurement timepoints, and the safety and efficacy outcomes assessed. Several trials identified significant improvements in cough frequency, severity, or health-related quality of life measures while participants were on treatment, although these improvements did not persist in any of the studies that included a post-treatment follow-up timepoint.

CONCLUSIONS

In the absence of an approved therapy, placebo remains the most common comparator in trials of potential RCC or UCC treatments. The between-study comparability of the published evidence is limited by heterogeneity of study design, study populations, and outcomes measures, as well as by concerns regarding study size and risk of bias.

Patient-reported experiences with refractory or unexplained chronic cough: a qualitative analysis

BACKGROUND

Chronic cough, defined as a cough lasting 8 or more weeks, affects up to 10% of adults. Refractory chronic cough (RCC) is a cough that is uncontrolled despite comprehensive investigation and treatment of comorbid conditions while unexplained chronic cough (UCC) is a cough with no identifiable cause despite extensive evaluation of comorbid conditions. RCC and UCC are often poorly controlled. Understanding individuals' lived experience of the symptoms and impacts of these conditions may guide therapeutic strategies.

OBJECTIVES

The primary objectives of this study were to assess respondents' perceptions of the key symptoms of RCC and UCC and the impacts of RCC and UCC and their symptoms on well-being, health-related quality of life, work productivity, and social relationships.

DESIGN

Qualitative study.

METHODS

This study enrolled 30 adults with physician-diagnosed RCC or UCC. Two trained qualitative researchers conducted individual, in-depth telephone interviews using a semi-structured interview guide. Interviews were audio-recorded, transcribed, coded, and systematically analyzed to identify content themes.

RESULTS

A total of 15 respondents with RCC and 15 with UCC were included in the study. Many respondents had RCC or UCC for a long duration (median 9 years, range: 0-24). Half of the respondents reported having a coughing episode at least once daily. Only 40% of respondents reported that medication had improved their symptoms. In over half of the respondents, RCC or UCC hindered communication, caused embarrassment, frustration, and worry, and lowered quality of life. Perceptions of meaningful treatment benefits in RCC or UCC varied widely across respondents.

CONCLUSION

RCC and UCC remained poorly managed in many individuals and were associated with a wide range of symptoms and cough triggers that hindered daily activities and reduced emotional well-being. Understanding individuals' lived experiences may inform the development of RCC and UCC therapeutic strategies.

Chronic cough: more than just a persistent cough: a systematic literature review to understand the impact of chronic cough on quality of life

PURPOSE

Chronic cough (CC), defined as a cough persisting ≥ 8 weeks, can have a substantial negative impact on health-related quality of life (HRQoL). This is exacerbated by challenges with timely diagnosis and a lack of approved therapies. A systematic literature review (SLR) was conducted to identify evidence on HRQoL and health state utility values associated with refractory CC or unexplained CC.

METHODS

Electronic database searches were supplemented with searches of conference proceedings and health technology assessment body websites. Two independent reviewers assessed all citations for inclusion based on predefined inclusion/exclusion criteria. Key inclusion criteria were patient populations with CC and reporting of patient-reported outcomes or utilities using generic or disease-specific measures.

RESULTS

Following screening, 65 studies were identified for inclusion in the SLR. Of these, 23 studies assessed HRQoL among patients with CC who were not treated or treated with unspecified interventions, and 42 studies in patients who were treated with specified interventions. The studies indicated a substantial decrement to HRQoL as a result of CC, characterized by generic and disease-specific patient-reported outcome measures. HRQoL was impacted across multiple domains, including physical, psychological, and social functioning. The studies also demonstrated the potential for treatments to have a significant positive impact on HRQoL.

CONCLUSIONS

CC can substantially affect HRQoL in patients, across physical, psychological, and social domains. Although treatments can improve HRQoL in these patients, the available evidence is limited. There remains an unmet need for approved pharmacological treatments to alleviate CC and improve HRQoL for these patients.

Understanding the economic burden of chronic cough: a systematic literature review

Chronic cough (CC) is associated with high healthcare resource utilization (HCRU) due to challenges in diagnosis and treatment and is anticipated to have a substantial economic impact. This systematic literature review (SLR) sought to identify evidence on the cost-effectiveness of treatments and the economic burden associated with CC. Electronic database searches were supplemented with searches of conference proceedings and health technology assessment body websites. Two independent reviewers assessed all citations for inclusion based on predefined inclusion/exclusion criteria. Key inclusion criteria were patient population with CC, and outcomes related to cost-effectiveness and HCRU and costs. After screening, one cost-effectiveness analysis was identified, alongside eight studies reporting HCRU and costs related to CC. Though evidence was limited, studies suggest that patients with CC incur higher costs and use more resources than those with acute cough. Types of resource use reported included healthcare contacts and prescriptions, diagnostic tests, referrals and specialist evaluations, and treatment use. There is a paucity of literature on HCRU and costs in CC, and very limited cost-effectiveness analyses. The economic burden appears higher in these patients however, without direct comparison to the general population it is difficult to determine the total impact. The increased burden is expected to be a result of the challenges with diagnosis and lack of approved treatments. However, limited conclusions can be drawn in the absence of further data. Future studies should endeavor to quantify the HCRU and cost attributable to patients with CC.

Health-related experiences of adults with chronic cough: Empirical research mixed methods

AIM

To describe adults' health-related experiences with chronic cough.

DESIGN

Survey and interviews.

METHODS

Participants completed questionnaires and interviews, to explore chronic cough's impact and management.

DATA SOURCES

Patients aged 18-85 years with at least three cough-related encounters within 56-120 days.

RESULTS

Forty-one patients were surveyed. Mean cough severity was 4.5 (scale 0-9). Chronic cough-related problems included embarrassment (66%), fatigue (56%), and anxiety or depression (49%). Testing was judged insufficient by 44%. Only 28% were satisfied with treatment; 20% reported abandoning treatment due to ineffectiveness. Interview themes (N = 30) included frustration with diagnostic uncertainty, and feelings of therapeutic futility. Some reported psychological distress. Work and socializing were commonly disrupted.

CONCLUSION

Diagnostic uncertainty, perceived limitations of testing, and treatment failures suggest needs for better approaches to evaluating and treating chronic cough. Special attention to identifying and addressing mental health issues appears warranted.

Prediction of Persistent Chronic Cough in Patients with Chronic Cough Using Machine Learning

Introduction

To develop and validate prediction models for risk of persistent chronic cough (PCC) in patients with chronic cough.

Study Design

Retrospective cohort study

Methods

Two retrospective cohorts of patients 18–85 years of age were identified for years 2011–2016: aspecialist cohort whichincluded CC patients diagnosed by specialists, and anevent cohortwhich comprised of CC patients identified by at least three cough events. A cough event could be a cough diagnosis, dispensing of cough medication, or any indication of cough in clinical notes. Model training and validation were conducted using two machine learning approaches and 400+ features. Sensitivity analyses were also conducted. PCC was defined as a CC diagnosis or any two (specialist cohort) or three (event cohort) cough events in year 2 and again in year 3 after the index date.

Results

8581/52 010 patients met eligibility criteria for the specialist/event cohorts (mean age 60.0/55.5 years), respectively. 38.2% and 12.4% patients in the specialist and event cohorts, respectively, developed PCC. The utilization-based models were mainly based on baseline health care utilizations associated with CC or respiratory diseases, while the diagnosis-based models incorporated traditional parameters including age, asthma, pulmonary fibrosis, obstructive pulmonary disease, gastroesophageal reflux, hypertension, and bronchiectasis. All final models were parsimonious (5–7 predictors) and moderately accurate (AUC: 0.74–0.76 for utilization-based models and 0.71 for diagnosis-based models).

Conclusions

The application of our risk prediction models may be used to identify high-risk PCC patients at any stage of the clinical testing/evaluation to facilitate decision making.

Development of a natural language processing algorithm to detect chronic cough in electronic health records

Background

Chronic cough (CC) is difficult to identify in electronic health records (EHRs) due to the lack of specific diagnostic codes. We developed a natural language processing (NLP) model to identify cough in free-text provider notes in EHRs from multiple health care providers with the objective of using the model in a rules-based CC algorithm to identify individuals with CC from EHRs and to describe the demographic and clinical characteristics of individuals with CC.

Methods

This was a retrospective observational study of enrollees in Optum’s Integrated Clinical + Claims Database. Participants were 18–85 years of age with medical and pharmacy health insurance coverage between January 2016 and March 2017. A labeled reference standard data set was constructed by manually annotating 1000 randomly selected provider notes from the EHRs of enrollees with ≥ 1 cough mention. An NLP model was developed to extract positive or negated cough contexts. NLP, cough diagnosis and medications identified cough encounters. Patients with ≥ 3 encounters spanning at least 56 days within 120 days were defined as having CC.

Results

The positive predictive value and sensitivity of the NLP algorithm were 0.96 and 0.68, respectively, for positive cough contexts, and 0.96 and 0.84, respectively, for negated cough contexts. Among the 4818 individuals identified as having CC, 37% were identified using NLP-identified cough mentions in provider notes alone, 16% by diagnosis codes and/or written medication orders, and 47% through a combination of provider notes and diagnosis codes/medications. Chronic cough patients were, on average, 61.0 years and 67.0% were female. The most prevalent comorbidities were respiratory infections (75%) and other lower respiratory disease (82%).

Conclusions

Our EHR-based algorithm integrating NLP methodology with structured fields was able to identify a CC population. Machine learning based approaches can therefore aid in patient selection for future CC research studies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12890-022-02035-6.

Risk Factors for Persistent Chronic Cough During Consecutive Years: A Retrospective Database Analysis

BACKGROUND

The identification of patients at high risk for diseases provides clinicians essential information to better manage such patients. Persistent chronic cough (PCC) is a condition with high clinical burden and limited knowledge of the risk factors that drive the persistent symptoms.

OBJECTIVE

To understand the risk factors of PCC in patients with CC diagnosed by specialists.

METHODS

In this retrospective study, adults aged 18 to 85 years diagnosed with CC by a pulmonologist, allergist, otolaryngologist, or gastroenterologist in the period 2011 to 2016 were identified. PCC was defined by another CC code or at least 2 cough events at least 8 weeks but no more than 4 months apart in each of the 2 consecutive years beginning 1 year after the original CC diagnosis. Unadjusted and adjusted risk ratios with 95% CI for patient characteristics at baseline in relationship to PCC were estimated by Poisson regression models with robust error variance.

RESULTS

Of the adults with CC, 3270 (27.4%) had PCC and 5302 (44.5%) did not have CC during follow-up; 3341 (28.1%) had CC in only 1 follow-up year and were excluded from the analysis. Compared with patients without PCC, patients with PCC were noted to have significantly increased adjusted risk ratios for the following baseline features: (1) demographic characteristics (elderly, females, and less educated); (2) comorbidities (chronic obstructive pulmonary disease, chronic sinusitis, bronchiectasis, pulmonary fibrosis, hypertension, depression, and cough complications); (3) medication dispensed (inhaled corticosteroids/long-acting beta-agonists, leukotriene modifiers, nasal corticosteroids, nasal short-acting muscarinic antagonists, proton pump inhibitors, antitussives with narcotics, and neuromodulators); and (4) specialist care, particularly with pulmonologists.

CONCLUSIONS

Knowledge of the independent risk factors associated with PCC should aid clinicians in identifying such patients and improve their management.

Application of unsupervised deep learning algorithms for identification of specific clusters of chronic cough patients from EMR data

BACKGROUND

Chronic cough affects approximately 10% of adults. The lack of ICD codes for chronic cough makes it challenging to apply supervised learning methods to predict the characteristics of chronic cough patients, thereby requiring the identification of chronic cough patients by other mechanisms. We developed a deep clustering algorithm with auto-encoder embedding (DCAE) to identify clusters of chronic cough patients based on data from a large cohort of 264,146 patients from the Electronic Medical Records (EMR) system. We constructed features using the diagnosis within the EMR, then built a clustering-oriented loss function directly on embedded features of the deep autoencoder to jointly perform feature refinement and cluster assignment. Lastly, we performed statistical analysis on the identified clusters to characterize the chronic cough patients compared to the non-chronic cough patients.

RESULTS

The experimental results show that the DCAE model generated three chronic cough clusters and one non-chronic cough patient cluster. We found various diagnoses, medications, and lab tests highly associated with chronic cough patients by comparing the chronic cough cluster with the non-chronic cough cluster. Comparison of chronic cough clusters demonstrated that certain combinations of medications and diagnoses characterize some chronic cough clusters.

CONCLUSIONS

To the best of our knowledge, this study is the first to test the potential of unsupervised deep learning methods for chronic cough investigation, which also shows a great advantage over existing algorithms for patient data clustering.

Applying interpretable deep learning models to identify chronic cough patients using EHR data

BACKGROUND AND OBJECTIVE

Chronic cough (CC) affects approximately 10% of adults. Many disease states are associated with chronic cough, such as asthma, upper airway cough syndrome, bronchitis, and gastroesophageal reflux disease. The lack of an ICD code specific for chronic cough makes it challenging to identify such patients from electronic health records (EHRs). For clinical and research purposes, computational methods using EHR data are urgently needed to identify chronic cough cases. This research aims to investigate the data representations and deep learning algorithms for chronic cough prediction.

METHODS

Utilizing real-world EHR data from a large academic healthcare system from October 2005 to September 2015, we investigated Natural Language Representation of the EHR data and systematically evaluated deep learning and traditional machine learning models to predict chronic cough patients. We built these machine learning models using structured data (medication and diagnosis) and unstructured data (clinical notes).

RESULTS

The sensitivity and specificity of a transformer-based deep learning algorithm, specifically BERT with attention model, was 0.856 and 0.866, respectively, using structured data (medication and diagnosis). Sensitivity and specificity improved to 0.952 and 0.930 when we combined structured data with symptoms extracted from clinical notes. We further found that the attention mechanism of deep learning models can be used to extract important features that drive the prediction decisions. Compared with our previously published rule-based algorithm, the deep learning algorithm can identify more chronic cough patients with structured data.

CONCLUSIONS

By applying deep learning models, chronic cough patients can be reliably identified for prospective or retrospective research through medication and diagnosis data, widely available in EHR and electronic claims data, thus improving the generalizability of the patient identification algorithm. Deep learning models can identify chronic cough patients with even higher sensitivity and specificity when structured and unstructured EHR data are utilized. We anticipate language-based data representation and deep learning models developed in this research could also be productively used for other disease prediction and case identification.

Patient-Reported Burden of Chronic Cough in a Managed Care Organization

BACKGROUND

The burden of chronic cough (CC) requires better understanding.

OBJECTIVE

To determine the severity, health status, and health care resource utilization among patients with CC identified by electronic health records on 2 visits separated by ≥1 year.

METHODS

Information on cough-related burden was collected through survey from patients with CC, including validated questionnaires (the cough health status Leicester Cough Questionnaire [LCQ], the cough hypersensitivity Hull Airway Reflux Questionnaire [HARQ], and the Cough Quality of Life Questionnaire [CQLQ]), CC-associated respiratory and gastrointestinal comorbidities, and treatment responses. Spearman correlation coefficients were reported to examine the associations among the LCQ, HARQ, and CQLQ. Patient demographics and patient-reported CC features were compared between males and females, and among ethnic groups using Robust Poisson regression models.

RESULTS

The survey was completed by 565 patients who were 64.8 ± 12.6 years, 75.8% female, and 60.4% white. CC duration was 8.6 ± 10.5 years with an average weekly severity of 5.3 ± 2.3 (maximum 10). The LCQ score was 11.3 ± 3.9 (maximum 21). The HARQ score was 33.3 ± 13.6 (normal ≤13). The CQLQ score was 56.9 ± 17.5 (maximum 112, worse with higher scores). The Spearman rank correlations were high between the LCQ and HARQ (-0.65), the LCQ and CQLQ (-0.80), and the HARQ and CQLQ (0.69). Patients with CC-associated respiratory and gastrointestinal comorbidities generally showed similar results regarding the above questionnaires. Treatment responses were suboptimal. Women compared with men and non-whites compared with whites reported significantly worse cough severity and poorer LCQ, HARQ, and CQLQ scores.

CONCLUSIONS

CC is self-reported as a burdensome condition, particularly in women and non-white minorities, which markedly affects daily living with inadequate response to treatments.

Burden of Specialist-Diagnosed Chronic Cough in Adults

BACKGROUND

The burden of chronic cough (CC) in patients seeking specialist care is infrequently researched.

OBJECTIVE

To describe patient characteristics and disease burden associated with specialist-diagnosed CC.

METHODS

Using administrative pharmacy and medical data, we identified patients aged 18 to 85 years with CC diagnosed by pulmonologists, allergists, otolaryngologists, or gastroenterologists. Patients were stratified into 4 subgroups on the basis of the presence or absence of common respiratory diseases or gastroesophageal reflux disease (GERD). Clinical features and health care resource utilization (HCRU) in the baseline and outcome years were compared between the CC subgroups. The baseline factors associated with persistence of CC and a comparison of the CC cohort to a matched noncough cohort were also determined.

RESULTS

The 11,290 patients with specialist-diagnosed CC were aged about 61 years and 66.7% were females. The CC cohort experienced frequent GERD (44.1%), asthma (31.2%), obesity (24.3%), upper airway cough syndrome (20.4%), common cough complications (19.4%), and hospitalizations (9.8%). The patients with CC with both respiratory disease and GERD exhibited at baseline and follow-up the most common cough comorbidities, higher HCRU, specialist care, and dispensed respiratory and nonrespiratory medications including proton pump inhibitors, antitussives, psychotherapeutics, oral corticosteroids, and antibiotics compared with the other subgroups. A 40.6% persistence of CC occurred similarly between CC subgroups. In addition, patients with CC in the matched analysis experienced significantly more comorbidities, laboratory evaluations, HCRU, and antitussives than patients without cough.

CONCLUSIONS

Specialist-diagnosed CC was associated with considerable disease burden, particularly among those with both respiratory disease and GERD. In addition, CC burden was more pronounced than in matched patients without cough.

Prevalence and Characteristics of Chronic Cough in Adults Identified by Administrative Data

CONTEXT

International Classification of Diseases-9/10 codes for chronic cough (CC) do not exist, limiting investigation.

OBJECTIVE

To develop a computerized algorithm to determine CC prevalence and its characteristics.

DESIGN

This observational study using administrative data identified hierarchically patients aged 18 to 85 years with CC from 2013 to 2016. First, a specialist-diagnosed CC group was identified using an internal CC encounter code during an outpatient visit to a pulmonologist, allergist, otolaryngologist, or gastroenterologist. Subsequently, an event-diagnosed CC group was identified based on clinical notes through natural language processing, ICD-9/ICD-10 cough codes, and dispensed antitussives.

MAIN OUTCOME MEASURES

Prevalence of CC and comparison of clinical characteristics between specialist-diagnosed and event-diagnosed CC subgroups.

RESULTS

A total of 50,163 patients with CC of more than 8 weeks were identified. Of these, 11,290 (22.5%) were specialist diagnosed, and 38,873 (77.5%) were event diagnosed. The CC cohort was 57.4 ± 16.5 years of age; 67.6% were female. The overall prevalence was 1.04% (95% confidence interval = 1.03-1.06) in 2016. Prevalence in 2016 was higher in female patients (1.21%) than in male patients (0.81%), higher in patients aged 65 to 85 years (2.2%) than in patients aged 18 to 44 years (0.43%), and higher in Blacks (1.38%) than in Whites (1.21%). Compared with patients with event-diagnosed CC, patients with specialist-diagnosed CC exhibited significantly higher frequencies of laboratory tests and respiratory and nonrespiratory comorbidities and dispensed medication and lower frequency of pneumonia, all-cause and respiratory-cause emergency department visits and hospitalizations, and dispensed antitussives.

CONCLUSIONS

We identified a CC cohort using electronic data in a managed care organization. Prevalences varied by sex, age, and ethnicity. Clinical characteristics varied between specialist-diagnosed and event-diagnosed CC.

Secondary prevention of diabetes through workplace health screening

Background

Workplace health screening offers a unique opportunity to assess individuals for type 2 diabetes mellitus.

Aims

To evaluate the association between workplace diabetes screening, subsequent diagnosis and changes in fasting plasma glucose (FPG), glycated haemoglobin (HbA1c) and body mass index (BMI) among individuals who screened positive for diabetes.

Methods

Employees without a prior diagnosis of diabetes participated in workplace health screening by 45 employers throughout the USA. Individuals screened positive for diabetes based on standard criteria (≥126 mg/dL FPG or ≥6.5% [48 mmol/mol] HbA1c). Diabetes diagnoses were identified after screening using claims-based ICD9-CM diagnosis codes. Discrete-time survival analysis estimated the monthly rate of new diabetes cases after screening, relative to the time period before screening. Paired t-tests evaluated 1-year changes in blood glucose measures and BMI among individuals with positive screenings.

Results

Of 22790 participating individuals, 900 (4%) screened positive for diabetes. A significantly greater rate of new diabetes diagnoses was observed during the first month after screening, compared to the 3-month period before screening (odds ratio [OR] 2.65, 95% confidence intervals [CIs] 2.02-3.47). Among 538 individuals with diabetes who returned for workplace screening 1 year later, significant improvements were observed in BMI (mean ± SD = -0.63 ± 2.56 kg/m2, P < 0.001) and FPG levels (mean ± SD = -9.3 ± 66.5 mg/dL, P < 0.01).

Conclusions

Workplace screening was associated with a reduction in the number of undiagnosed employees with diabetes and significant improvement in FPG and BMI at 1-year follow-up.

Anticholinergic burden and risk of cognitive impairment in elderly nursing home residents with depression

BACKGROUND

Although the adverse cognitive effects of anticholinergic medications in the elderly are well-documented, little is known regarding the cognitive impact of anticholinergics among nursing home residents with depression.

OBJECTIVE

This study examined the risk of mild-to-moderate cognitive impairment due to anticholinergic burden among elderly nursing home residents with depression.

METHODS

A population-based nested case-control study was conducted using Minimum Data Set (MDS)-linked Medicare data where the base cohort included patients ≥ 65 years with depression who had intact cognition (MDS Cognition score of 0 or 1) and no dementia. Cases were identified as those who had mild-to-moderate cognition (MDS Cognition score of 2-4). Each case was matched on age and sex to one control using incidence density sampling. The study evaluated cumulative anticholinergic burden (defined as score of 3 or more) within 30, 60 and 90 days preceding the event date based on the Anticholinergic Drug Scale (ADS). Conditional logistic regression model stratified on matched case-control sets was performed to evalaute cognitive impairment due to cumulative anticholinergic burden after controlling for other risk factors.

RESULTS

The study sample included 3707 cases with mild-to-moderate cognition and 3707 matched controls with intact cognition. Bivariate analysis showed significant association between cumulative anticholinergic exposure and cognitive impairment (Odds Ratio [OR], 1.15; 95% Confidence Interval [CI],1.04-1.30); after controlling for potential risk factors, cumulative anticholinergic exposure 30 days preceding the event was no longer associated with cognitive impairment, (aOR, 1.07; 95% CI, 0.95-1.21). However, the odds of cognitive impairment increased with cumulative anticholinergic exposure 60 days (aOR 1.16; 1.04-1.30) and 90 days (aOR 1.28; 1.14-1.44) before the event date.

CONCLUSION

Cumulative anticholinergic use for prolonged exposure periods was associated with modestly increased risk of cognitive impairment in elderly residents with depression who had intact cognition. The findings suggest the need to be cautious when prescribing multiple anticholinergic drugs in residents, including those with intact cognition.

Cardiovascular Safety of Concomitant Use of Atypical Antipsychotics and Long-Acting Stimulants in Children and Adolescents With ADHD

OBJECTIVE

This study examined cardiovascular safety of concomitant use of long-acting stimulants (LAS) and atypical antipsychotics (AAP) in children and adolescents with Attention Deficit Hyperactivity Disorder (ADHD).

METHOD

The study used 2004-2007 IMS LifeLink™ claims data involving 6- to 16-year-old children with ADHD and at least one LAS prescription from July 2004 to December 2006. Time-dependent Cox regression analysis was performed to evaluate the risk of cardiovascular disease (CVD) events due to concomitant use of LAS and AAP.

RESULTS

The analytical cohort consisted of 37,903 children: 538 (1.9%) used LAS and AAP concurrently and the rest used LAS monotherapy. Time-dependent Cox regression analysis revealed no difference in CVD risk among concomitant users of LAS and AAP (hazard ratio = 1.19; 95% confidence interval = [0.60, 2.53]) when compared with users of LAS monotherapy.

CONCLUSION

Concomitant use of LAS and AAP was not associated with risk of CVD events in ADHD patients when compared with LAS monotherapy.

Functional Status of Elderly Nursing Home Residents With Parkinson's Disease

BACKGROUND

Parkinson's disease (PD) is a debilitating neurological disorder that has a significant impact on activities of daily living (ADL). Limited data exists regarding the functional status of nursing home residents with PD.

OBJECTIVE

The study examined the nature, extent, and predictors of limitation of ADL in patients with PD residing in nursing homes.

METHODS

This study used cross-sectional design involving data from the National Nursing Home Survey (NNHS). The dependent variable was the total number of ADL for which the patient needed assistance. Assistance for ADL comprised of five items of transferring, dressing, eating, toileting and bathing. Multivariable linear regression was used to examine the individual, biological and environmental factors associated with assistance needed for ADL.

RESULTS

The nationally representative study sample was of 98,093 nursing home residents with PD for an overall prevalence of 6.57%. The mean age of the study sample was 81.03±0.34 years. Mood symptoms (45%), behavioral symptoms (24%)and dementia (8%) were common among PD patients. Assistance needed for each ADL ranged from 68.24% for eating to 99.25% for bathing. Older age, being married, living in metropolitan statistical area (MSA), living in other place prior to admission, and presence of mood symptoms were positively related to assistance for ADL. Presence of behavioral symptoms and use of levodopa combination therapy were negatively related to assistance for ADL.

CONCLUSIONS

Majority of the nursing home residents with PD had disability in all the five items of ADL. Various individual, environmental and biological factors were associated with ADL in nursing home residents with PD.

Risk of Cognitive Decline Associated With Paroxetine Use in Elderly Nursing Home Patients With Depression

OBJECTIVE

This study evaluated the risk of cognitive decline associated with paroxetine use in elderly nursing home patients with depression.

METHODS

A retrospective cohort study was conducted using the 2007 to 2010 Medicare Part D claims and minimum data set (MDS) data involving new users of paroxetine and other selective serotonin reuptake inhibitors (SSRIs). The primary outcome was MDS Cognition Scale. The repeated-measures mixed model was used to examine the effect of paroxetine on cognition after controlling for other factors.

RESULTS

The baseline MDS Cognition Scale measures for paroxetine (n = 63) and other SSRI users (n =1018) were 2.02 (±1.85) and 2.50 (±2.39), respectively. The repeated-measures mixed model did not find statistically significant difference in cognition with the use of paroxetine (β = 0.02, 95% CI: -0.16 to 0.21]) when compared to other SSRIs.

CONCLUSIONS

There was no differential effect of paroxetine on cognition when compared to other SSRIs.

Identification of Undiagnosed Hyperlipidemia: Do Work Site Screening Programs Work?

PURPOSE

We evaluated the rate of hyperlipidemia identified during workplace screening in previously undiagnosed individuals, the association between workplace hyperlipidemia screening and use of medical care during follow-up, and changes in lipid profile among individuals with hyperlipidemia at screening.

DESIGN

Nonexperimental longitudinal study.

SETTING

Employees who participated in a workplace health screening.

PARTICIPANTS

A total of 18 993 individuals from 39 self-insured employers in the United States.

MEASURES

Total cholesterol, low-density lipoprotein (LDL), high-density lipoprotein (HDL), and triglycerides were measured during screening. A claims-based algorithm was used to identify hyperlipidemia cases.

ANALYSIS

Discrete-time survival analysis was used to estimate monthly rates of new hyperlipidemia diagnoses or prescriptions. Paired t tests were used to evaluate 1-year changes in lipid profile.

RESULTS

A total of 1872 (9.9%) individuals had hyperlipidemia at screening. Among all individuals, a significantly greater rate of new hyperlipidemia diagnoses was observed during the first month after screening, compared to the 3 months before screening (odds ratio [95% CI]: 2.99 [2.66-3.36]). Among the 987 individuals who were followed up 1 year later, significant improvements were observed in total cholesterol (-8.5% ± 13.6%) and LDL levels (-10.2% ± 19.3%).

CONCLUSION

Workplace health screenings in an insured population were associated with a subsequent increase in physician visits and prescriptions for hyperlipidemia. After 1 year, significant improvements in total cholesterol and LDL levels were observed among individuals who screened positive for hyperlipidemia.

Comparative risk of hip fractures in elderly nursing home patients with depression using paroxetine and other selective serotonin reuptake inhibitors

AIM

To evaluate comparative safety of paroxetine and other selective serotonin reuptake inhibitors (SSRIs) for the risk of hip fractures.

PATIENTS & METHODS

A propensity score-matched retrospective cohort study was conducted using 2007-2010 Minimum Data Set linked Medicare data. Robust Cox proportional hazards model was used to evaluate the risk of hip fractures in depressed elderly nursing home residents.

RESULTS

Cox analysis did not find any significant difference in the risk of hip fractures for the paroxetine users (hazard ratio: 1.09; 95% CI: 0.91-1.32) when compared with other SSRIs. Results from the sensitivity analysis supported the main findings.

CONCLUSION

There was no differential risk of hip fractures between paroxetine and other SSRIs. Future studies are needed to evaluate other anticholinergic effects of paroxetine.

Risk of Dementia Among Elderly Nursing Home Patients Using Paroxetine and Other Selective Serotonin Reuptake Inhibitors

OBJECTIVE

Selective serotonin reuptake inhibitors (SSRIs) are the first line of treatment for depression. Among the SSRIs, paroxetine has strong anticholinergic properties and may lead to increased risk of adverse cognitive outcomes among elderly patients. This study evaluated the comparative risk of dementia associated with use of paroxetine and other SSRIs among elderly nursing home patients.

METHODS

A retrospective cohort study using propensity score matching was conducted with 2007-2010 Minimum Data Set-linked Medicare data. The study population included elderly nursing home patients with depression who were new users of SSRIs. Patients were followed for a maximum of two years after index SSRI use. The risk of dementia was modeled by using a robust Cox proportional hazards model to account for clustering within matched users of paroxetine and other SSRIs.

RESULTS

The unmatched cohort included 19,952 new users of SSRIs; 1,898 used paroxetine, and 18,054 used other SSRIs. In the propensity-matched cohort of 3,796 patients, the unadjusted incidence of dementia was 7.5% for users of paroxetine and 8.6% for users of other SSRIs. There was no difference in the risk of dementia for users of paroxetine or other SSRIs. These study findings remained robust in multiple sensitivity analyses involving various measures of dementia.

CONCLUSIONS

Compared with use of other SSRIs, use of paroxetine was not associated with higher risk of dementia among elderly nursing home patients with depression. Future studies are needed to evaluate the impact of paroxetine on other cognition measures.

Comparative Cognitive Profile of Second-Generation Antidepressants in Elderly Nursing Home Residents With Depression

BACKGROUND

Past literature suggests that the use of second-generation antidepressants improves cognition in depressed elderly patients.

OBJECTIVE

This study assessed the comparative cognitive profile of commonly used second-generation antidepressant classes in elderly residents with depression.

METHODS

A multiple propensity score adjusted retrospective cohort study was conducted using 2007-2010 Medicare Part D claims and Minimum Data Set (MDS). Elderly nursing home residents (65 years or older) with depression using a new prescription of selective serotonin reuptake inhibitors (SSRIs), serotonin-norepinephrine reuptake inhibitors (SNRIs), and tetracyclics constituted the study cohort. The outcome of interest was cognition, measured using the MDS Cognition Scale. Cognition was measured at each quarterly assessment after antidepressant initiation for a maximum of 1 year. The propensity score-adjusted repeated-measures mixed model was used to evaluate the comparative profile of SSRIs, SNRIs, and tetracyclics with respect to cognition.

RESULTS

The study cohort comprised 1518 elderly nursing home residents. Of these, 1081 received SSRIs (71.21%), 320 received tetracyclics (21.08%), and 117 received SNRIs (7.71%). The propensity score-adjusted repeated-measures mixed model did not show any statistically significant difference in cognition with the use of SSRIs (β = -0.14; 95% CI = -0.53, 0.25) or tetracyclics (β = -0.36; 95% CI = -0.80, 0.08) when compared with SNRIs, after controlling for other factors.

CONCLUSIONS

The cognitive effect of SSRIs, SNRIs, and tetracyclics was similar in elderly nursing home residents with depression. Further studies are needed to evaluate the long-term cognitive effects of second-generation antidepressants in this vulnerable population.

Novel screening metric for the identification of at-risk peripheral artery disease patients using administrative claims data

Despite high morbidity and mortality associated with peripheral artery disease (PAD), it remains under-diagnosed and under-treated. The objective of this study was to develop a screening metric to identify undiagnosed patients at high risk of developing PAD using administrative data. Commercial claims data from 2010 to 2012 were utilized to develop and internally validate a PAD screening metric. Medicare data were used for external validation. The study population included adults, aged 30 years or older, with new cases of PAD identified using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis/procedure codes or the Healthcare Common Procedure Coding System (HCPCS) codes. Multivariate logistic regression was conducted to determine PAD risk factors used in the development of the screening metric for the identification of at-risk PAD patients. The cumulative incidence of PAD was 6.6%. Sex, age, congestive heart failure, hypertension, chronic renal insufficiency, stroke, diabetes, acute myocardial infarction, transient ischemic attack, hyperlipidemia, and angina were significant risk factors for PAD. A cut-off score of ⩾20 yielded sensitivity, specificity, positive predictive value, negative predictive value, and c-statistics of 83.5%, 60.0%, 12.8%, 98.1%, and 0.78, respectively. By identifying patients at high risk for developing PAD using only administrative data, the use of the current pre-screening metric could reduce the number of diagnostic tests, while still capturing those patients with undiagnosed PAD.

Predictors of concomitant use of antipsychotics and stimulants and its impact on stimulant persistence in pediatric attention deficit hyperactivity disorder

BACKGROUND

Concomitant use of stimulants and atypical antipsychotics is common in pediatric attention deficit hyperactivity disorder (ADHD). However, little is known about the determinants of concomitant use and its utility in the management of pediatric ADHD.

OBJECTIVES

To (a) examine predictors of concomitant stimulant and atypical antipsychotic use and (b) evaluate the impact of concomitant atypical antipsychotic use on the persistence of stimulants in children and adolescents diagnosed with ADHD.

METHODS

The retrospective cohort study was conducted using 4 years (January 2004-December 2007) of IMS LifeLink claims data. The study population included children and adolescents aged 6-16 years with a diagnosis of ADHD and those who initiated long-acting stimulants (LAS) from July 2004 to December 2006. Patients were followed for 1 year after index stimulant use. Concomitant use was defined as the concurrent prescription for LAS and atypical antipsychotic agents with at least 14 days overlap after the index LAS claim. Persistence was measured by summing the total number of days a patient remained on the index LAS from the index prescription date with an allowable gap of no more than 30 days. Multiple logistic regression within the conceptual framework of the Andersen Behavioral Model was performed to determine the predictors of concomitant stimulant and atypical antipsychotic use. Multivariate Cox proportional hazards regression within the conceptual framework of the Andersen Behavioral Model was used to examine the impact of concomitant atypical antipsychotic use on persistence of stimulants.

RESULTS

The study cohort consisted of 39,981 children who initiated LAS treatment. Most (96.10%) received LAS monotherapy, and 3.90% received LAS and atypical antipsychotic concomitantly. The multiple logistic regression analysis found that gender, health insurance, region, year of cohort entry, season, physician specialty, coexisting mental health conditions, and general mental health status influenced the concomitant use of LAS and atypical antipsychotic agents. Bivariate analyses revealed that concomitant users had longer persistence (by 71 days) than the stimulant-alone users. Cox proportional hazards regression revealed that concomitant atypical antipsychotic was associated with improvement in LAS persistence by 15% (HR = 0.85, 95% CI = 0.76-0.94) in comparison with the LAS recipients who did not use atypical antipsychotic concomitantly. Other factors such as age, region, season, coexisting mental health conditions, use of comedications, and general mental health status influenced the LAS treatment persistence among children and adolescents.

CONCLUSIONS

Various predisposing, enabling, and need factors were associated with the concomitant stimulant and atypical antipsychotic use. Concomitant use of atypical antipsychotics was associated with improved LAS treatment persistence in children and adolescents with ADHD.

Which risk-adjustment index performs better in predicting 30-day mortality? A systematic review and meta-analysis

RATIONALE, AIMS AND OBJECTIVES

Individual comparisons of the performance of risk-adjustment indices have been widely conducted. Few reviews have been conducted to summarize the performance of different risk-adjustment indices. A 30-day mortality rate is widely used to evaluate the quality of care in hospitals by federal agencies like the Centers for Medicare and Medicaid Services. This study examined relative performance of risk-adjustment indices that predict 30-day mortality.

METHODS

Databases including Medline, PubMed and PsycINFO were searched for studies that compared risk-adjustment indices. The search protocol included comparative studies in which the performance of risk-adjustment indices were compared across any defined cohort to compare 30-day mortality, including mortality within 30 days and intensive care unit mortality, which lasts less than 30 days. Data were extracted using a structured form and abstract data included author and publication year, population studied (including location, sample size, study time period), comparison indices, outcome studied, results and conclusions from the results. A meta-analytical approach was used to summarize all the studies. Scaled ranking score was used to estimate the relative superiority of any given risk-adjustment indices. A hypergeometric test was carried out to evaluate the performance of risk-adjustment measures.

RESULTS

Out of 2805 studies identified, 23 studies met the eligibility criteria. Main risk-adjustment indices used for comparison included Acute Physiology and Chronic Health Evaluation, Sequential Organ Failure Assessment score, Charlson co-morbidity index, Model for End-Stage Liver Disease score and Simplified Acute Physiology Score (SAPS). Based on scaled ranking score, SAPS performed best (score 0.510) among all the risk-adjustment indices. However, based on hypergeometric test, the five measures performed equally well.

CONCLUSIONS

Although all the selected risk-adjustment indices perform equally well, SAPS seems better than other indices for short-term mortality based on scaled ranking score.

Determinants of nonmedical use, abuse or dependence on prescription drugs, and use of substance abuse treatment

BACKGROUND

Previous studies have found a negative association between health insurance and nonmedical use of prescription drugs (NMUPD), and abuse or dependence on prescription drugs (ADPD); and mixed associations between health insurance and use of substance abuse treatment (SAT). However, effect of health insurance in the specific subgroups of population is largely unknown.

OBJECTIVE

To estimate the relationship between health insurance and (1) NMUPD, (2) ADPD, and (3) use of SAT services among 12-64 years old, noninstitutionalized individuals and to see if these relationships are different in different subgroups of population.

METHODS

This study used cross-sectional survey data from 2007 National Survey on Drug Use and Health. Bivariate and multiple logistic regression analyses were conducted.

RESULTS

In 2007, self-reported prevalence of NMUPD was approximately 10% (N=15,509,703). In multivariate analysis, NMUPD was negatively associated with health insurance, age, race other than non-Hispanic White, education, marital status, and income ($40,000-$74,999). Past year use of tobacco and alcohol were positively associated with NMUPD. Among those with private health insurance, Hispanics and individuals with family income less than $20,000 and $40,000-$74,999 were more likely prone to NMUPD than others. High school graduates with public health insurance were less likely prone to NMUPD. Approximately, 13% of nonmedical users reported ADPD (N=2,011,229). Health insurance and age were negatively associated with ADPD. However, people who were unmarried, reported fair/poor health, and used tobacco were more likely to report ADPD. Lastly, the use of substance abuse treatment programs was approximately 73% and 76% between NMUPD and ADPD population, respectively. Health insurance was not associated with use of substance abuse treatment. Individuals with high school education were 2.6 times more likely to use substance abuse treatment than the college graduates. Additionally, no significant interaction effects were found between health insurance, and sociodemographic factors on ADPD and the use of substance abuse treatment.

CONCLUSIONS

Health insurance had a differential impact on NMUPD only. Among privately insured, Hispanics and individuals reporting family income less than $20,000 were more likely to engage in NMUPD. There is a need to better understand and monitor the use of prescription drugs among these groups. This knowledge can help in developing public health programs and policies that discourage NMUPD among these individuals.