Development and Validation of a Predictive Model to Identify Individuals Likely to Have Undiagnosed Chronic Obstructive Pulmonary Disease Using an Administrative Claims Database

Enabling chronic obstructive pulmonary disease diagnosis through chest X-rays: A multi-site and multi-modality study

PURPOSE

Chronic obstructive pulmonary disease (COPD) is one of the most common chronic illnesses in the world. Unfortunately, COPD is often difficult to diagnose early when interventions can alter the disease course, and it is underdiagnosed or only diagnosed too late for effective treatment. Currently, spirometry is the gold standard for diagnosing COPD but it can be challenging to obtain, especially in resource-poor countries. Chest X-rays (CXRs), however, are readily available and may have the potential as a screening tool to identify patients with COPD who should undergo further testing or intervention. In this study, we used three CXR datasets alongside their respective electronic health records (EHR) to develop and externally validate our models.

METHOD

To leverage the performance of convolutional neural network models, we proposed two fusion schemes: (1) model-level fusion, using Bootstrap aggregating to aggregate predictions from two models, (2) data-level fusion, using CXR image data from different institutions or multi-modal data, CXR image data, and EHR data for model training. Fairness analysis was then performed to evaluate the models across different demographic groups.

RESULTS

Our results demonstrate that DL models can detect COPD using CXRs with an area under the curve of over 0.75, which could facilitate patient screening for COPD, especially in low-resource regions where CXRs are more accessible than spirometry.

CONCLUSIONS

By using a ubiquitous test, future research could build on this work to detect COPD in patients early who would not otherwise have been diagnosed or treated, altering the course of this highly morbid disease.

Distance-Based Detection of Cough, Wheeze, and Breath Sounds on Wearable Devices

Smart wearable sensors are essential for continuous health-monitoring applications and detection accuracy of symptoms and energy efficiency of processing algorithms are key challenges for such devices. While several machine-learning-based algorithms for the detection of abnormal breath sounds are reported in literature, they are either too computationally expensive to implement into a wearable device or inaccurate in multi-class detection. In this paper, a kernel-like minimum distance classifier (K-MDC) for acoustic signal processing in wearable devices was proposed. The proposed algorithm was tested with data acquired from open-source databases, participants, and hospitals. It was observed that the proposed K-MDC classifier achieves accurate detection in up to 91.23% of cases, and it reaches various detection accuracies with a fewer number of features compared with other classifiers. The proposed algorithm's low computational complexity and classification effectiveness translate to great potential for implementation in health-monitoring wearable devices.

Spirometry evaluation to assess performance of a claims-based predictive model identifying patients with undiagnosed COPD

Background

A claims-based model to predict patients likely to have undiagnosed COPD was developed by Moretz et al in 2015. This study aims to assess the performance of the aforementioned model using prospectively collected spirometry data.

Methods

A study population aged 40–89 years enrolled in a Medicare Advantage plan with prescription drug coverage or commercial health plan and without a claim for COPD diagnosis was identified from April 1, 2012 to March 31, 2016 in the Humana claims database. This population was stratified into subjects likely or unlikely to have undiagnosed COPD using the claims-based predictive model. Subjects were randomly selected for spirometry evaluation of FEV₁ and FVC. The predictive model was validated using airflow limitation ratio (FEV₁/FVC <0.70).

Results

A total of 218 subjects classified by the predictive model as likely and 331 not likely to have undiagnosed COPD completed spirometry evaluation. Those predicted to have undiagnosed COPD had a higher mean age (70.2 vs 67.9 years, P=0.0012) and a lower mean FEV₁/FVC ratio (0.724 vs 0.753, P=0.0002) compared to those predicted not to have undiagnosed COPD. Performance metrics for the predictive model were: area under the curve =0.61, sensitivity =52.5%, specificity =64.6%, positive predictive value =33.5%, and negative predictive value =80.1%.

Conclusion

The claims-based predictive model identifies those not at risk of having COPD eight out of ten times, and those who are likely to have COPD one out of three times.

Development of a claims-based algorithm to identify potentially undiagnosed chronic migraine patients

OBJECTIVE

To develop a claims-based algorithm to identify undiagnosed chronic migraine among patients enrolled in a healthcare system.

METHODS

An observational study using claims and patient survey data was conducted in a large medical group. Eligible patients had an International Classification of Diseases, Ninth/Tenth Revision (ICD-9/10) migraine diagnosis, without a chronic migraine diagnosis, in the 12 months before screening and did not have a migraine-related onabotulinumtoxinA claim in the 12 months before enrollment. Trained clinicians administered a semi-structured diagnostic interview, which served as the gold standard to diagnose chronic migraine, to enrolled patients. Potential claims-based predictors of chronic migraine that differentiated semi-structured diagnostic interview-positive (chronic migraine) and semi-structured diagnostic interview-negative (non-chronic migraine) patients were identified in bivariate analyses for inclusion in a logistic regression model.

RESULTS

The final sample included 108 patients (chronic migraine = 64; non-chronic migraine = 44). Four significant predictors for chronic migraine were identified using claims in the 12 months before enrollment: ≥15 versus <15 claims for acute treatment of migraine, including opioids (odds ratio = 5.87 [95% confidence interval: 1.34-25.63]); ≥24 versus <24 healthcare visits (odds ratio = 2.80 [confidence interval: 1.08-7.25]); female versus male sex (odds ratio = 9.17 [confidence interval: 1.26-66.50); claims for ≥2 versus 0 unique migraine preventive classes (odds ratio = 4.39 [confidence interval: 1.19-16.22]). Model sensitivity was 78.1%; specificity was 72.7%.

CONCLUSIONS

The claims-based algorithm identified undiagnosed chronic migraine with sufficient sensitivity and specificity to have potential utility as a chronic migraine case-finding tool using health claims data. Research to further validate the algorithm is recommended.

Predicting Acute Exacerbations in Chronic Obstructive Pulmonary Disease

BACKGROUND

With increasing health care costs that have outpaced those of other industries, payers of health care are moving from a fee-for-service payment model to one in which reimbursement is tied to outcomes. Chronic obstructive pulmonary disease (COPD) is a disease where this payment model has been implemented by some payers, and COPD exacerbations are a quality metric that is used. Under an outcomes-based payment model, it is important for health systems to be able to identify patients at risk for poor outcomes so that they can target interventions to improve outcomes.

OBJECTIVE

To develop and evaluate predictive models that could be used to identify patients at high risk for COPD exacerbations.

METHODS

This study was retrospective and observational and included COPD patients treated with a bronchodilator-based combination therapy. We used health insurance claims data to obtain demographics, enrollment information, comorbidities, medication use, and health care resource utilization for each patient over a 6-month baseline period. Exacerbations were examined over a 6-month outcome period and included inpatient (primary discharge diagnosis for COPD), outpatient, and emergency department (outpatient/emergency department visits with a COPD diagnosis plus an acute prescription for an antibiotic or corticosteroid within 5 days) exacerbations. The cohort was split into training (75%) and validation (25%) sets. Within the training cohort, stepwise logistic regression models were created to evaluate risk of exacerbations based on factors measured during the baseline period. Models were evaluated using sensitivity, specificity, and positive and negative predictive values. The base model included all confounding or effect modifier covariates. Several other models were explored using different sets of observations and variables to determine the best predictive model.

RESULTS

There were 478,772 patients included in the analytic sample, of which 40.5% had exacerbations during the outcome period. Patients with exacerbations had slightly more comorbidities, medication use, and health care resource utilization compared with patients without exacerbations. In the base model, sensitivity was 41.6% and specificity was 85.5%. Positive and negative predictive values were 66.2% and 68.2%, respectively. Other models that were evaluated resulted in similar test characteristics as the base model.

CONCLUSIONS

In this study, we were not able to predict COPD exacerbations with a high level of accuracy using health insurance claims data from COPD patients treated with bronchodilator-based combination therapy. Future studies should be done to explore predictive models for exacerbations.

DISCLOSURES

No outside funding supported this study. Samp is now employed by, and owns stock in, AbbVie. The other authors have nothing to disclose. Study concept and design were contributed by Joo and Pickard, along with the other authors. Samp and Lee performed the data analysis, with assistance from the other authors. Samp wrote the manuscript, which was revised by Schumock and Calip, along with the other authors.

Development and validation of a predictive model to identify patients at risk of severe COPD exacerbations using administrative claims data

Background

Patients with COPD often experience severe exacerbations involving hospitalization, which accelerate lung function decline and reduce quality of life. This study aimed to develop and validate a predictive model to identify patients at risk of developing severe COPD exacerbations using administrative claims data, to facilitate appropriate disease management programs.

Methods

A predictive model was developed using a retrospective cohort of COPD patients aged 55–89 years identified between July 1, 2010 and June 30, 2013 using Humana’s claims data. The baseline period was 12 months postdiagnosis, and the prediction period covered months 12–24. Patients with and without severe exacerbations in the prediction period were compared to identify characteristics associated with severe COPD exacerbations. Models were developed using stepwise logistic regression, and a final model was chosen to optimize sensitivity, specificity, positive predictive value (PPV), and negative PV (NPV).

Results

Of 45,722 patients, 5,317 had severe exacerbations in the prediction period. Patients with severe exacerbations had significantly higher comorbidity burden, use of respiratory medications, and tobacco-cessation counseling compared to those without severe exacerbations in the baseline period. The predictive model included 29 variables that were significantly associated with severe exacerbations. The strongest predictors were prior severe exacerbations and higher Deyo–Charlson comorbidity score (OR 1.50 and 1.47, respectively). The best-performing predictive model had an area under the curve of 0.77. A receiver operating characteristic cutoff of 0.4 was chosen to optimize PPV, and the model had sensitivity of 17%, specificity of 98%, PPV of 48%, and NPV of 90%.

Conclusion

This study found that of every two patients identified by the predictive model to be at risk of severe exacerbation, one patient may have a severe exacerbation. Once at-risk patients are identified, appropriate maintenance medication, implementation of disease-management programs, and education may prevent future exacerbations.

A Population-Based Study on Myelodysplastic Syndromes in the Lazio Region (Italy), Medical Miscoding and 11-Year Mortality Follow-Up: the Gruppo Romano-Laziale Mielodisplasie Experience of Retrospective Multicentric Registry

Data on Myelodysplastic Syndromes (MDS) are difficult to collect by cancer registries because of the lack of reporting and the use of different classifications of the disease. In the Lazio Region, data from patients with a confirmed diagnosis of MDS, treated by a hematology center, have been collected since 2002 by the Gruppo Romano-Laziale Mielodisplasie (GROM-L) registry, the second MDS registry existing in Italy. This study aimed at evaluating MDS medical miscoding during hospitalizations, and patients' survival. For these purposes, we selected 644 MDS patients enrolled in the GROM-L registry. This cohort was linked with two regional health information systems: the Hospital Information System (HIS) and the Mortality Information System (MIS) in the 2002-2012 period. Of the 442 patients who were hospitalized at least once during the study period, 92% had up to 12 hospitalizations. 28.5% of patients had no hospitalization episodes scored like MDS, code 238.7 of the International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM). The rate of death during a median follow-up of 46 months (range 0.9-130) was 45.5%. Acute myeloid leukemia (AML) was the first cause of mortality, interestingly a relevant portion of deaths is due to cerebro-cardiovascular events and second tumors. This study highlights that MDS diagnosis and treatment, which require considerable healthcare resources, tend to be under-documented in the HIS archive. Thus we need to improve the HIS to better identify information on MDS hospitalizations and outcome. Moreover, we underline the importance of comorbidity in MDS patients' survival.

Design of Wearable Breathing Sound Monitoring System for Real-Time Wheeze Detection

In the clinic, the wheezing sound is usually considered as an indicator symptom to reflect the degree of airway obstruction. The auscultation approach is the most common way to diagnose wheezing sounds, but it subjectively depends on the experience of the physician. Several previous studies attempted to extract the features of breathing sounds to detect wheezing sounds automatically. However, there is still a lack of suitable monitoring systems for real-time wheeze detection in daily life. In this study, a wearable and wireless breathing sound monitoring system for real-time wheeze detection was proposed. Moreover, a breathing sounds analysis algorithm was designed to continuously extract and analyze the features of breathing sounds to provide the objectively quantitative information of breathing sounds to professional physicians. Here, normalized spectral integration (NSI) was also designed and applied in wheeze detection. The proposed algorithm required only short-term data of breathing sounds and lower computational complexity to perform real-time wheeze detection, and is suitable to be implemented in a commercial portable device, which contains relatively low computing power and memory. From the experimental results, the proposed system could provide good performance on wheeze detection exactly and might be a useful assisting tool for analysis of breathing sounds in clinical diagnosis.