Procedure code overutilization detection from healthcare claims using unsupervised deep learning methods.

BACKGROUND: Fraud, Waste, and Abuse (FWA) in medical claims have a negative impact on the quality and cost of healthcare. A major component of FWA in claims is procedure code overutilization, where one or more prescribed procedures may not be relevant to a given diagnosis and patient profile, resulting in unnecessary and unwarranted treatments and medical payments. This study aims to identify such unwarranted procedures from millions of healthcare claims. In the absence of labeled examples of unwarranted procedures, the study focused on the application of unsupervised machine learning techniques. METHODS: Experiments were conducted with deep autoencoders to find claims containing anomalous procedure codes indicative of FWA, and were compared against a baseline density-based clustering model. Diagnoses, procedures, and demographic data associated with healthcare claims were used as features for the models. A dataset of one hundred thousand claims sampled from a larger claims database is used to initially train and tune the models, followed by experimentations on a dataset with thirty-three million claims. Experimental results show that the autoencoder model, when trained with a novel feature-weighted loss function, outperforms the density-based clustering approach in finding potential outlier procedure codes. RESULTS: Given the unsupervised nature of our experiments, model performance was evaluated using a synthetic outlier test dataset, and a manually annotated outlier test dataset. Precision, recall and F1-scores on the synthetic outlier test dataset for the autoencoder model trained on one hundred thousand claims were 0.87, 1.0 and 0.93, respectively, while the results for these metrics on the manually annotated outlier test dataset were 0.36, 0.86 and 0.51, respectively. The model performance on the manually annotated outlier test dataset improved further when trained on the larger thirty-three million claims dataset with precision, recall and F1-scores of 0.48, 0.90 and 0.63, respectively. CONCLUSIONS: This study demonstrates the feasibility of leveraging unsupervised, deep-learning methods to identify potential procedure overutilization from healthcare claims.

The PATRIC Bioinformatics Resource Center: expanding data and analysis capabilities.

The PathoSystems Resource Integration Center (PATRIC) is the bacterial Bioinformatics Resource Center funded by the National Institute of Allergy and Infectious Diseases (https://www.patricbrc.org). PATRIC supports bioinformatic analyses of all bacteria with a special emphasis on pathogens, offering a rich comparative analysis environment that provides users with access to over 250 000 uniformly annotated and publicly available genomes with curated metadata. PATRIC offers web-based visualization and comparative analysis tools, a private workspace in which users can analyze their own data in the context of the public collections, services that streamline complex bioinformatic workflows and command-line tools for bulk data analysis. Over the past several years, as genomic and other omics-related experiments have become more cost-effective and widespread, we have observed considerable growth in the usage of and demand for easy-to-use, publicly available bioinformatic tools and services. Here we report the recent updates to the PATRIC resource, including new web-based comparative analysis tools, eight new services and the release of a command-line interface to access, query and analyze data.

CD4 count outperforms World Health Organization clinical algorithm for point-of-care HIV diagnosis among hospitalised HIV-exposed Malawian infants.

OBJECTIVE: To determine, for the WHO algorithm for point-of-care diagnosis of HIV infection, the agreement levels between paediatricians and non-physician clinicians, and to compare sensitivity and specificity profiles of the WHO algorithm and different CD4 thresholds against HIV PCR testing in hospitalised Malawian infants. METHODS: In 2011, hospitalised HIV-exposed infants <12 months in Lilongwe, Malawi, were evaluated independently with the WHO algorithm by both a paediatrician and clinical officer. Blood was collected for CD4 and molecular HIV testing (DNA or RNA PCR). Using molecular testing as the reference, sensitivity, specificity and positive predictive value (PPV) were determined for the WHO algorithm and CD4 count thresholds of 1500 and 2000 cells/mm(3) by paediatricians and clinical officers. RESULTS: We enrolled 166 infants (50% female, 34% <2 months, 37% HIV infected). Sensitivity was higher using CD4 thresholds (<1500, 80%; <2000, 95%) than with the algorithm (physicians, 57%; clinical officers, 71%). Specificity was comparable for CD4 thresholds (<1500, 68%, <2000, 50%) and the algorithm (paediatricians, 55%, clinical officers, 50%). The positive predictive values were slightly better using CD4 thresholds (<1500, 59%, <2000, 52%) than the algorithm (paediatricians, 43%, clinical officers 45%) at this prevalence. CONCLUSION: Performance by the WHO algorithm and CD4 thresholds resulted in many misclassifications. Point-of-care CD4 thresholds of <1500 cells/mm(3) or <2000 cells/mm(3) could identify more HIV-infected infants with fewer false positives than the algorithm. However, a point-of-care option with better performance characteristics is needed for accurate, timely HIV diagnosis.

Empyema due to severe Pasteurella multocida treated with quinolone: a case report.

Pasteurella multocida (PM) is gram-negative coccobacillus that is commonly acquired through a cat scratch or bite. The standard antibiotic of choice for treating PM is penicillin. We present a case of PM empyema in a penicillin allergic 56-year-old female who has a history of contact with domestic cats and is known to have chronic obstructive pulmonary disease and a chronic history of smoking. The patient was treated successfully with Levofloxacin as alternative treatment to penicillin.

Development of a severity of illness scoring system (inpatient triage, assessment and treatment) for resource-constrained hospitals in developing countries.

OBJECTIVE: To develop a new paediatric illness severity score, called inpatient triage, assessment and treatment (ITAT), for resource-limited settings to identify hospitalised patients at highest risk of death and facilitate urgent clinical re-evaluation. METHODS: We performed a nested case-control study at a Malawian referral hospital. The ITAT score was derived from four equally weighted variables, yielding a cumulative score between 0 and 8. Variables included oxygen saturation, temperature, and age-adjusted heart and respiratory rates. We compared the ITAT score between cases (deaths) and controls (discharges) in predicting death within 2 days. Our analysis includes predictive statistics, bivariable and multivariable logistic regression, and calculation of data-driven scores. RESULTS: A total of 54 cases and 161 controls were included in the analysis. The area under the receiver operating characteristic curve was 0.76. At an ITAT cut-off of 4, the sensitivity, specificity and likelihood ratio were 0.44, 0.86 and 1.70, respectively. A cumulative ITAT score of 4 or higher was associated with increased odds of death (OR 4.80; 95% CI 2.39-9.64). A score of 2 for all individual vital signs was a statistically significant independent predictor of death. CONCLUSIONS: We developed an inpatient triage tool (ITAT) appropriate for resource-constrained hospitals that identifies high-risk children after hospital admission. Further research is needed to study how best to operationalise ITAT in developing countries.

Leveraging deep survival models to predict quality of care risk in diverse hospital readmissions.

Hospital readmissions rate is reportedly high and has caused huge financial burden on health care systems in many countries. It is viewed as an important indicator of health care providers' quality of care. We examine the use of machine learning-based survival analysis to assess quality of care risk in hospital readmissions. This study applies various survival models to explore the risk of hospital readmissions given patient demographics and their respective hospital discharges extracted from a health care claims dataset. We explore advanced feature representation techniques such as BioBERT and Node2Vec to encode high-dimensional diagnosis code features. To our knowledge, this study is the first to apply deep-learning based survival-analysis models for predicting hospital readmission risk agnostic of specific medical conditions and a fixed window for readmission. We found that modeling the time from discharge date to readmission date as a Weibull distribution as in the SparseDeepWeiSurv model yields the best discriminative power and calibration. In addition, embedding representations of the diagnosis codes do not contribute to improvement in model performance. We find dependency of each model's performance on the time point at which it is evaluated. This time dependency of the models' performance on the health care claims data may necessitate a different choice of model in quality of care issue detection at different points in time. We show the effectiveness of deep-learning based survival-analysis models in estimating the quality of care risk in hospital readmissions.

Listening panel agreement and characteristics of lung sounds digitally recorded from children aged 1-59 months enrolled in the Pneumonia Etiology Research for Child Health (PERCH) case-control study.

INTRODUCTION: Paediatric lung sound recordings can be systematically assessed, but methodological feasibility and validity is unknown, especially from developing countries. We examined the performance of acoustically interpreting recorded paediatric lung sounds and compared sound characteristics between cases and controls. METHODS: Pneumonia Etiology Research for Child Health staff in six African and Asian sites recorded lung sounds with a digital stethoscope in cases and controls. Cases aged 1-59 months had WHO severe or very severe pneumonia; age-matched community controls did not. A listening panel assigned examination results of normal, crackle, wheeze, crackle and wheeze or uninterpretable, with adjudication of discordant interpretations. Classifications were recategorised into any crackle, any wheeze or abnormal (any crackle or wheeze) and primary listener agreement (first two listeners) was analysed among interpretable examinations using the prevalence-adjusted, bias-adjusted kappa (PABAK). We examined predictors of disagreement with logistic regression and compared case and control lung sounds with descriptive statistics. RESULTS: Primary listeners considered 89.5% of 792 case and 92.4% of 301 control recordings interpretable. Among interpretable recordings, listeners agreed on the presence or absence of any abnormality in 74.9% (PABAK 0.50) of cases and 69.8% (PABAK 0.40) of controls, presence/absence of crackles in 70.6% (PABAK 0.41) of cases and 82.4% (PABAK 0.65) of controls and presence/absence of wheeze in 72.6% (PABAK 0.45) of cases and 73.8% (PABAK 0.48) of controls. Controls, tachypnoea, >3 uninterpretable chest positions, crying, upper airway noises and study site predicted listener disagreement. Among all interpretable examinations, 38.0% of cases and 84.9% of controls were normal (p<0.0001); wheezing was the most common sound (49.9%) in cases. CONCLUSIONS: Listening panel and case-control data suggests our methodology is feasible, likely valid and that small airway inflammation is common in WHO pneumonia. Digital auscultation may be an important future pneumonia diagnostic in developing countries.

A new way to teach an old topic: the cadaver-based anatomy short course for high school students.

An outreach program has been developed at the Department of Biological Sciences, Northern Illinois University (NIU), that offers an intensive short course in human gross anatomy to high school students and their teachers. Prior to the short course, the high school teachers attend a laboratory orientation, view the facilities, and then develop a syllabus suitable for the class needs. When the students arrive, they spend between 1 and 2 days rotating through five different work stations. Using their syllabus, they identify and learn the function of various structures on human cadavers and models. Evaluation and content testing of 143 students demonstrated a significant degree of enthusiasm and retention of the syllabus material at the end of the course. All teachers who participated in the program indicated that they wish to return the following year and new groups have already applied to attend. The results of the program demonstrate ways that universities can generate excitement about science and learning at the high school level.

Comparing the quality of passively-scattered proton and photon tomotherapy plans for brain and head and neck disease sites.

We compare the quality of photon IMRT (helical tomotherapy) with classic proton plans for brain, head and neck tumors, in terms of target dose uniformity and conformity along with organ-at-risk (OAR) sparing. Plans were created for twelve target volumes among eight cases. All patients were originally planned and treated using helical tomotherapy. Proton plans were generated using a passively-scattered beam model with a maximum range of 32 g cm(-2) (225 MeV), range modulation in 0.5 g cm(-2) increments and range compensators with 4.8 mm milling tool diameters. All proton plans were limited to two to four beams. Plan quality was compared using uniformity index (UI), conformation number (CN) and a EUD-based plan quality index (fEUD). For 11 of the 12 targets, UI was improved for the proton plan; on average, UI was 1.05 for protons versus 1.08 for tomotherapy. For 7 of the 12 targets, the tomotherapy plan exhibited more favorable CN. For proximal OARs, the improved dose conformity to the target volume from tomotherapy led to a lower maximum dose. For distal OARs, the maximum dose was much lower for proton plans. For 6 of the 8 cases, near-total avoidance for distal OARs provided by protons leads to improved fEUD. However, if distal OARs are excluded in the fEUD calculation, the proton plans exhibit better fEUD in only 3 of the 8 cases. The distal OAR sparing and target dose uniformity are generally better with passive-scatter proton planning than with photon tomotherapy; proton therapy may be preferred if the clinician deems those attributes critical. However, tomotherapy may serve equally as well as protons for cases where superior target dose conformity from tomotherapy leads to plan quality nearly identical to or better than protons and for cases where distal OAR sparing is not concerning.

The c-myc x E2F-1/p21 interactive gene expression index augments cytomorphologic diagnosis of lung cancer in fine-needle aspirate specimens.

Morphological analysis of cytologic samples obtained by fine-needle aspirate (FNA) or bronchoscopy is an important method for diagnosing bronchogenic carcinoma. However, this approach has only about 65 to 80% diagnostic sensitivity. Based on previous studies, the c-myc x E2F-1/p21WAF1/CIP1 (p21 hereafter) gene expression index is highly sensitive and specific for distinguishing normal from malignant bronchial epithelial tissues. In an effort to improve sensitivity of diagnosing lung cancer in cytologic specimens, we used Standardized Reverse Transcriptase Polymerase Chain Reaction (StaRT-PCR) to measure the c-myc x E2F-1/p21 index in cDNA samples from 14 normal lung samples (6 normal lung parenchyma and 8 normal bronchial epithelial cell [NBEC] biopsies), and 16 FNA biopsies from 14 suspected tumors. Based on cytomorphologic criteria, 11 of the 14 suspected tumors were diagnosed as bronchogenic carcinoma and three specimens were non-diagnostic. Subsequent biopsy samples confirmed that the three non-diagnostic samples were derived from lung carcinomas. The index value for each bronchogenic carcinoma was above a cut-off value of 7000 and the index value of all but one normal sample was below 7000. Thus the c-myc x E2F-1/p21 index may augment cytomorphologic diagnosis of bronchogenic carcinoma biopsy samples, particularly those considered non-diagnostic by cytomorphologic criteria.

Determining the quality of IMCI pneumonia care in Malawian children.

BACKGROUND: Although pneumonia is the leading cause of child mortality worldwide, little is known about the quality of routine pneumonia care in high burden settings like Malawi that utilize World Health Organization's Integrated Management of Childhood Illnesses (IMCI) guidelines. Due to severe human resource constraints, the majority of clinical care in Malawi is delivered by non-physician clinicians called Clinical Officers (COs). AIM: To assess the quality of child pneumonia care delivered by Malawian COs in routine care conditions. METHODS: At an outpatient district-level clinic in Lilongwe, Malawi, 10 COs caring for 695 children who presented with fever, cough, or difficulty breathing were compared to IMCI pneumonia diagnostic and treatment guidelines. RESULTS: Fewer than 1% of patients received an evaluation by COs that included all 16 elements of the history and physical examination. The respiratory rate was only determined in 16.1% of patients presenting with cough or difficulty breathing. Of the 274 children with IMCI-defined pneumonia, COs correctly diagnosed 30%, and administered correct pneumonia care in less than 25%. COs failed to hospitalize 40.8% of children with severe or very severe pneumonia. CONCLUSIONS: IMCI pneumonia care quality at this Malawian government clinic is alarmingly low. Along with reassessing current pneumonia training and supervision approaches, novel quality improvement interventions are necessary to improve care.

Clinical versus rapid molecular HIV diagnosis in hospitalized African infants: a randomized controlled trial simulating point-of-care infant testing.

OBJECTIVE: Many African infants fail to receive their diagnostic HIV molecular test results and subsequently, antiretroviral therapy (ART). To determine whether a point-of-care molecular HIV test increases ART access for hospitalized Malawian infants, we simulated a point-of-care test using rapid HIV RNA polymerase chain reaction (Rapid PCR) and compared patient outcomes with an optimized standard care that included assessment with the World Health Organization clinical algorithm for HIV infection plus a DNA PCR with a turnaround time of several weeks (standard care). DESIGN: Randomized controlled trial. METHODS: Hospitalized HIV-exposed Malawian infants aged <12 months were randomized into Rapid PCR or standard care. Rapid PCR infants obtained molecular test results within 48 hours to facilitate immediate ART, similar to a point-of-care test. Standard care infants meeting clinical criteria were also offered inpatient ART. The primary outcome was appropriate in-hospital ART for DNA or RNA PCR-confirmed HIV-infected infants. RESULTS: Three hundred infants were enrolled. A greater proportion of HIV-infected infants receiving Rapid PCR, versus standard care, started inpatient ART (72.3% vs 47.8%, P = 0.016). Among molecular test-negative infants, 26.9% receiving standard care unnecessarily initiated inpatient ART, versus 0.0% receiving Rapid PCR (P < 0.001). Rapid PCR modestly reduced the median days to ART (3.0 vs 6.5, P = 0.001) but did not influence outpatient follow-up for HIV-infected infants (78.1% vs 82.4%, P = 0.418). CONCLUSIONS: Rapid PCR, versus an optimized standard care, increased the proportion of hospitalized HIV-infected infants initiating ART and reduced ART exposure in molecular test-negative infants, without meaningfully impacting time to ART initiation or follow-up rates.

Superior uptake and outcomes of early infant diagnosis of HIV services at an immunization clinic versus an "under-five" general pediatric clinic in Malawi.

OBJECTIVE: Although the Malawian government recommends HIV-exposed infants receive early infant diagnosis (EID) of HIV at "under-five" pediatric clinics (U5Cs), most never enroll. Therefore, we evaluated the integration of EID testing into an immunization clinic (IC) compared with the current standard of EID testing at an U5C. DESIGN: Prospective observational study. METHODS: Using routine provider-initiated HIV testing and counseling (PITC) registers, we prospectively studied 1757 children offered PITC at a government IC and U5C. Infants tested by HIV DNA polymerase chain reaction (PCR) were followed until PCR result disclosure or defaulting. RESULTS: We sampled 877 and 880 consecutive PITC recipients at U5C and IC, respectively. Overall, a 7-fold greater proportion received PITC at IC (84.2% vs. 11.4%, P < 0.001). PITC recipients at IC were more than 14 months younger (2.6 vs. 17.0, P < 0.001), with greater proportions HIV exposed (17.6% vs. 5.3%, P < 0.001) and PCR eligible (7.9% vs. 3.5%, P < 0.001). A higher percentage of IC infants accepted PCR testing (100.0% vs. 90.3%, P = 0.03). Additionally, IC PCR recipients were 2.5 months younger (3.1 vs. 5.6, P < 0.001) with 4 times less testing PCR positive (7.1% vs. 32.1%, P < 0.001). Importantly, a more than 3-fold greater proportion of HIV-exposed infants at IC returned for their PCR result and enrolled into care (78.6% vs. 25.0%, P < 0.001). CONCLUSIONS: Compared with an U5C, integrating EID testing into an IC is more acceptable, more feasible, enrolls more infants into EID at younger ages, and would likely strengthen Malawi's EID services if expanded.

Outcomes in children treated for perineal group A beta-hemolytic streptococcal dermatitis.

OBJECTIVES: To evaluate reports that describe relapse or recurrence following treatment of perineal streptococcal dermatitis (PSD), we studied a large cohort of children with these perianal or perivaginal infections to determine whether outcomes are related to the antimicrobial agent selected for initial treatment. METHODS: We audited laboratory logs and medical records to retrospectively identify incident cases of culture-confirmed PSD in children at a large university-affiliated health system during 2006-2008. We estimated rates of recurrence (defined as any return visit with a clinical diagnosis of perineal dermatitis within 6 months) and, then, incorporated these rates into a meta-analysis that included 8 previous studies. RESULTS: A total of 81 children had incident PSD during the study period, and 26 (32.1%) had a recurrence. Most (18/26 [69.2%]) had their first recurrence within 6 weeks. Among children treated with an oral agent, the recurrence rate was 16/42 (38.1%) following penicillin or amoxicillin and 10/36 (27.8%) following a beta-lactamase resistant agent (adjusted odds ratio: 2.02 [95% confidence interval {CI}: 0.69-5.92]). In the meta-analysis, recurrence rates following penicillin or amoxicillin were consistent across studies (fixed-effect test for heterogeneity, P = 0.35), and the pooled rate (37.4% [95% CI: 28.8%-46.5%]) was higher than observed following a beta-lactamase resistant agent (odds ratio: 2.39 [95% CI: 1.18-4.81]). CONCLUSIONS: Perineal streptococcal dermatitis initially treated with penicillin or amoxicillin is consistently associated with a high risk of clinical recurrence. Whether treatment with a beta-lactamase resistant agent reduces this risk is uncertain and should be subjected to a clinical trial.

Disseminated blastomycosis.

A 26-year-old veiled Saudi-Arabian woman presented with hemoptysis, and multiple nodules and abscesses. A skin biopsy specimen revealed yeast forms consistent with Blastomyces dermatitidis. Fungal cultures from bronchoscopy and skin specimens also grew B dermatitidis. She was treated with oral itraconazole (200 mg twice a day). Both lung and skin lesions showed improvement within 6 weeks.