Predicting cardiovascular disease risk using photoplethysmography and deep learning.

Cardiovascular diseases (CVDs) are responsible for a large proportion of premature deaths in low- and middle-income countries. Early CVD detection and intervention is critical in these populations, yet many existing CVD risk scores require a physical examination or lab measurements, which can be challenging in such health systems due to limited accessibility. We investigated the potential to use photoplethysmography (PPG), a sensing technology available on most smartphones that can potentially enable large-scale screening at low cost, for CVD risk prediction. We developed a deep learning PPG-based CVD risk score (DLS) to predict the probability of having major adverse cardiovascular events (MACE: non-fatal myocardial infarction, stroke, and cardiovascular death) within ten years, given only age, sex, smoking status and PPG as predictors. We compare the DLS with the office-based refit-WHO score, which adopts the shared predictors from WHO and Globorisk scores (age, sex, smoking status, height, weight and systolic blood pressure) but refitted on the UK Biobank (UKB) cohort. All models were trained on a development dataset (141,509 participants) and evaluated on a geographically separate test (54,856 participants) dataset, both from UKB. DLS's C-statistic (71.1%, 95% CI 69.9-72.4) is non-inferior to office-based refit-WHO score (70.9%, 95% CI 69.7-72.2; non-inferiority margin of 2.5%, p<0.01) in the test dataset. The calibration of the DLS is satisfactory, with a 1.8% mean absolute calibration error. Adding DLS features to the office-based score increases the C-statistic by 1.0% (95% CI 0.6-1.4). DLS predicts ten-year MACE risk comparable with the office-based refit-WHO score. Interpretability analyses suggest that the DLS-extracted features are related to PPG waveform morphology and are independent of heart rate. Our study provides a proof-of-concept and suggests the potential of a PPG-based approach strategies for community-based primary prevention in resource-limited regions.

Traumatic cervical spine fracture patterns on CT: a retrospective analysis at a level 1 trauma center.

PURPOSE: The purpose of our study was to determine common acute traumatic cervical spine fracture patterns on CT cervical spine (CTCS). METHODS: We retrospectively reviewed 1091 CTCS positive for traumatic fractures performed over a 10-year period at a level 1 trauma center. Fractures were classified by vertebral level, laterality, and anatomic location (anterior/posterior arch, body, odontoid, pedicle, facet, lateral mass, lamina, spinous process, transverse foramina, and transverse processes). RESULTS: C2 was the most commonly fractured vertebra (38% of all studies), followed by C7 (32.4%). 48.7% of studies had upper cervical spine (C1 and/or C2) fractures. 39.7% of positive studies involved > 1 vertebral level. Conditioned on fractures at one cervical level, the probability of fracture was greatest at adjacent levels with a 50% chance of sustaining a C7 fracture with C6 fracture. However, 31.3% (136) of studies with multi-level fractures had non-contiguous fractures. The most common isolated vertebral process fracture was of the transverse process, seen in 89 (8.2%) studies at a single level, 27 (2.5%) studies at multiple levels. Subaxial spine vertebral process fractures outnumbered body fractures with progressive dominance of vertebral process fracture down the spine. CONCLUSION: C2 was the most commonly fractured vertebral level. Multi-level traumatic cervical spine fractures constituted 40% of our cohort, most commonly at C6/C7 and C1/C2. Although the conditional probability of concurrent fracture in studies with multi-level fractures was greatest in contiguous levels, nearly one-third of multi-level fractures involved non-contiguous fractures.

Radiological and clinical features of traumatic atlanto-occipital dislocation.

PURPOSE: We aimed to describe the findings of traumatic atlanto-occipital dislocation (AOD) on cervical spine CTs and differences leading to varying treatment of these patients. METHODS: We retrospectively identified 20 adult patients with AOD from cervical spine CTs demonstrating fracture or fracture dislocations over 19 years at 2 major trauma centers. Medical records were reviewed and craniovertebral junction (CVJ) metrics measured on CT. Intubation, Glasgow Coma Scale (GCS), additional injuries, occiput/atlas/axis fracture, concurrent atlantoaxial subluxation, vascular injury on CT angiography, and ligamentous injury on MRI were noted. RESULTS: Using the Traynelis Classification, eight patients had type 2 and eight patients type 3 AOD. Four of 5 patients who died within 14 days of CT had type 2 AOD. Three patients had medial/lateral AOD. Of the patients who survived initial injuries, a greater percentage who underwent surgical or halo fixation versus non-operatively treated patients had abnormal CVJ measurements including BDI (62.5% vs 0%), atlantoaxial subluxation (75% vs 14.3%), ligamentous injury (80% vs 66.7%), intubation (62.5% vs 28.6%), GCS<8 (62.5% vs 14.3%), and additional injuries (75% vs 71.4%) on presentation. MRI helped identify 2 cases of type 2 AOD and surgical decision making in 8 cases. CONCLUSIONS: Types 2 and 3 were the most common, and type 2 is the deadliest type of AOD. A greater proportion of patients who undergo surgical or halo fixation have abnormal CT/MR findings with neurologic impairment at presentation. MRI aided detection of potentially missed type 2 AOD and was critical for surgical decision making.

Best practices for authors of healthcare-related artificial intelligence manuscripts.

Since its inception in 2017, npj Digital Medicine has attracted a disproportionate number of manuscripts reporting on uses of artificial intelligence. This field has matured rapidly in the past several years. There was initial fascination with the algorithms themselves (machine learning, deep learning, convoluted neural networks) and the use of these algorithms to make predictions that often surpassed prevailing benchmarks. As the discipline has matured, individuals have called attention to aberrancies in the output of these algorithms. In particular, criticisms have been widely circulated that algorithmically developed models may have limited generalizability due to overfitting to the training data and may systematically perpetuate various forms of biases inherent in the training data, including race, gender, age, and health state or fitness level (Challen et al. BMJ Qual. Saf. 28:231-237, 2019; O'neil. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy, Broadway Book, 2016). Given our interest in publishing the highest quality papers and the growing volume of submissions using AI algorithms, we offer a list of criteria that authors should consider before submitting papers to npj Digital Medicine.

Association between numeracy and self-rated poor health in 33 high- and upper middle- income countries.

The association between numeracy proficiency and health outcomes has been the subject of several studies. However, it is not known if this association is independent of educational attainment and literacy proficiency. In this study, we used logistic regression to model numeracy proficiency as a predictor of self-rated poor health after accounting for educational attainment and literacy proficiency. The prevalence of self-rated poor health among 166,863 adults aged 16-65 years from 33 high- and upper middle-income countries was 24%. Compared to those with the highest numeracy proficiency (level 4), the odds ratio of self-rated poor health for those with the lowest numeracy proficiency (level 1) was 2.2 (95% CI 1.9-2.7) and attenuated to 1.8 (95% CI 1.5-2.1) and 1.5 (95% CI 1.1, 2.0), respectively, after sequential addition of self-education and literacy proficiency. For those who were assessed to have low levels of both numeracy and literacy proficiency, the odds ratio of self-rated poor health was 1.4 (95% CI 1.3 to 1.5), relative to those who had high levels of both numeracy and literacy proficiencies. Numeracy and literacy proficiencies show both independent and interdependent correlations with poor self-rated health. Further, these associations varied by sociodemographic characteristics and across countries. Policies aimed at improving numeracy and literacy may be beneficial in preventing adverse health outcomes.

A prospective study of causes of death in rural Gadchiroli, an underdeveloped district of India (2011-2013).

BACKGROUND: Reliable information on causes of death to understand health priorities is rare from rural underdeveloped regions of India but is needed to direct health care response. This prompted us to study causes of death in a rural region of Gadchiroli, one of the most underdeveloped districts of India. METHODS: Data on causes of death were collected from 86 villages between April 2011 and March 2013 using verbal autopsies. Two physicians independently assigned cause of death using the tenth revision of the International Classification of Disease and disagreement was resolved by a third physician. RESULTS: There were 1599 deaths over 188,308 person years of observation. The crude death rate was 8.5 (95% confidence interval (CI)=8.1-8.9). The five leading causes of death were diseases of the circulatory system (20.8%), with stroke being the leading cause (14.3%), infections and parasitic disorders (18.4%), injuries and other external causes of mortality (10%), peri-natal diseases (6.5%) and diseases of the respiratory system (6.4%). Non-communicable diseases (NCDs) accounted for 38.5%, infections and perinatal diseases for 28.3% and external causes including injuries for 10% of all deaths. CONCLUSIONS: Epidemiological transition with a shift in causes of deaths from communicable to NCDs has occurred even in a rural underdeveloped district like Gadchiroli. Public health system in rural India which focuses on infections and maternal and child health will now need to be strengthened urgently to face the triple challenge of communicable and non-communicable diseases as well as injuries.

Validating a Machine Learning Algorithm to Predict 30-Day Re-Admissions in Patients With Heart Failure: Protocol for a Prospective Cohort Study.

BACKGROUND: Big data solutions, particularly machine learning predictive algorithms, have demonstrated the ability to unlock value from data in real time in many settings outside of health care. Rapid growth in electronic medical record adoption and the shift from a volume-based to a value-based reimbursement structure in the US health care system has spurred investments in machine learning solutions. Machine learning methods can be used to build flexible, customized, and automated predictive models to optimize resource allocation and improve the efficiency and quality of health care. However, these models are prone to the problems of overfitting, confounding, and decay in predictive performance over time. It is, therefore, necessary to evaluate machine learning-based predictive models in an independent dataset before they can be adopted in the clinical practice. In this paper, we describe the protocol for independent, prospective validation of a machine learning-based model trained to predict the risk of 30-day re-admission in patients with heart failure. OBJECTIVE: This study aims to prospectively validate a machine learning-based predictive model for inpatient admissions in patients with heart failure by comparing its predictions of risk for 30-day re-admissions against outcomes observed prospectively in an independent patient cohort. METHODS: All adult patients with heart failure who are discharged alive from an inpatient admission will be prospectively monitored for 30-day re-admissions through reports generated by the electronic medical record system. Of these, patients who are part of the training dataset will be excluded to avoid information leakage to the algorithm. An expected sample size of 1228 index admissions will be required to observe a minimum of 100 30-day re-admission events. Deidentified structured and unstructured data will be fed to the algorithm, and its prediction will be recorded. The overall model performance will be assessed using the concordance statistic. Furthermore, multiple discrimination thresholds for screening high-risk patients will be evaluated according to the sensitivity, specificity, predictive values, and estimated cost savings to our health care system. RESULTS: The project received funding in April 2017 and data collection began in June 2017. Enrollment was completed in July 2017. Data analysis is currently underway, and the first results are expected to be submitted for publication in October 2018. CONCLUSIONS: To the best of our knowledge, this is one of the first studies to prospectively evaluate a predictive machine learning algorithm in a real-world setting. Findings from this study will help to measure the robustness of predictions made by machine learning algorithms and set a realistic benchmark for expectations of gains that can be made through its application to health care. REGISTERED REPORT IDENTIFIER: RR1-10.2196/9466.

A machine learning model to predict the risk of 30-day readmissions in patients with heart failure: a retrospective analysis of electronic medical records data.

BACKGROUND: Heart failure is one of the leading causes of hospitalization in the United States. Advances in big data solutions allow for storage, management, and mining of large volumes of structured and semi-structured data, such as complex healthcare data. Applying these advances to complex healthcare data has led to the development of risk prediction models to help identify patients who would benefit most from disease management programs in an effort to reduce readmissions and healthcare cost, but the results of these efforts have been varied. The primary aim of this study was to develop a 30-day readmission risk prediction model for heart failure patients discharged from a hospital admission. METHODS: We used longitudinal electronic medical record data of heart failure patients admitted within a large healthcare system. Feature vectors included structured demographic, utilization, and clinical data, as well as selected extracts of un-structured data from clinician-authored notes. The risk prediction model was developed using deep unified networks (DUNs), a new mesh-like network structure of deep learning designed to avoid over-fitting. The model was validated with 10-fold cross-validation and results compared to models based on logistic regression, gradient boosting, and maxout networks. Overall model performance was assessed using concordance statistic. We also selected a discrimination threshold based on maximum projected cost saving to the Partners Healthcare system. RESULTS: Data from 11,510 patients with 27,334 admissions and 6369 30-day readmissions were used to train the model. After data processing, the final model included 3512 variables. The DUNs model had the best performance after 10-fold cross-validation. AUCs for prediction models were 0.664 ± 0.015, 0.650 ± 0.011, 0.695 ± 0.016 and 0.705 ± 0.015 for logistic regression, gradient boosting, maxout networks, and DUNs respectively. The DUNs model had an accuracy of 76.4% at the classification threshold that corresponded with maximum cost saving to the hospital. CONCLUSIONS: Deep learning techniques performed better than other traditional techniques in developing this EMR-based prediction model for 30-day readmissions in heart failure patients. Such models can be used to identify heart failure patients with impending hospitalization, enabling care teams to target interventions at their most high-risk patients and improving overall clinical outcomes.

Assessing the Usability of an Automated Continuous Temperature Monitoring Device (iThermonitor) in Pediatric Patients: Non-Randomized Pilot Study.

BACKGROUND: Fever is an important vital sign and often the first one to be assessed in a sick child. In acutely ill children, caregivers are expected to monitor a child's body temperature at home after an initial medical consult. Fever literacy of many caregivers is known to be poor, leading to fever phobia. In children with a serious illness, the responsibility of periodically monitoring temperature can add substantially to the already stressful experience of caring for a sick child. OBJECTIVE: The objective of this pilot study was to assess the feasibility of using the iThermonitor, an automated temperature measurement device, for continuous temperature monitoring in postoperative and postchemotherapy pediatric patients. METHODS: We recruited 25 patient-caregiver dyads from the Pediatric Surgery Department at the Massachusetts General Hospital (MGH) and the Pediatric Cancer Centers at the MGH and the Dana Farber Cancer Institute. Enrolled dyads were asked to use the iThermonitor device for continuous temperature monitoring over a 2-week period. Surveys were administered to caregivers at enrollment and at study closeout. Caregivers were also asked to complete a daily event-monitoring log. The Generalized Anxiety Disorder-7 item questionnaire was also used to assess caregiver anxiety at enrollment and closeout. RESULTS: Overall, 19 participant dyads completed the study. All 19 caregivers reported to have viewed temperature data on the study-provided iPad tablet at least once per day, and more than a third caregivers did so six or more times per day. Of all participants, 74% (14/19) reported experiencing an out-of-range temperature alert at least once during the study. Majority of caregivers reported that it was easy to learn how to use the device and that they felt confident about monitoring their child's temperature with it. Only 21% (4/9) of caregivers reported concurrently using a device other than the iThermonitor to monitor their child's temperature during the study. Continuous temperature monitoring was not associated with an increase in caregiver anxiety. CONCLUSIONS: The study results reveal that the iThermonitor is a highly feasible and easy-to-use device for continuous temperature monitoring in pediatric oncology and surgery patients. TRIAL REGISTRATION: ClinicalTrials.gov NCT02410252; https://clinicaltrials.gov/ct2/show/NCT02410252 (Archived by WebCite at http://www.webcitation.org/73LnO7hel).

Association between literacy and self-rated poor health in 33 high- and upper middle-income countries.

OBJECTIVES: To assess the relationship between general literacy proficiency and self-rated poor health by analyzing data from the Programme for the International Assessment of Adult Competencies, an international survey conducted from 2011 to 2015 in 33 high- and upper middle-income countries and national sub-regions. METHODS: Logistic regression was used to model general literacy proficiency as a predictor of self-rated poor health. RESULTS: Data from 167,062 adults aged 25-65 years were analyzed. The mean overall prevalence of self-rated poor health was 24%. The odds ratio of self-rated poor health for those in the lowest level of general literacy proficiency compared to those in the highest level was 2.5 (95% CI 2.2-3.0) in the unadjusted model, and 1.9 (95% CI 1.6-2.2) in the adjusted model. This association was robust over time and across countries. General literacy proficiency attenuated 22% of the effect of self-education on self-rated poor health, in addition to a substantial independent effect of its own. CONCLUSIONS: Our study provides robust and generalizable evidence that general literacy proficiency is independently associated with self-rated poor health. These results offer a potential modifiable target for policy interventions to reduce educational inequities in health.

Dietary determinants of serum total cholesterol among middle-aged and older adults: a population-based cross-sectional study in Dar es Salaam, Tanzania.

OBJECTIVE: To assess the dietary determinants of serum total cholesterol. DESIGN: Cross-sectional population-based study. SETTING: Peri-urban region of Dar es Salaam, Tanzania. PARTICIPANTS: 347 adults aged 40 years and older from the Dar es Salaam Urban Cohort Hypertension Study. MAIN OUTCOME MEASURE: Serum total cholesterol measured using a point-of-care device. RESULTS: Mean serum total cholesterol level was 204 mg/dL (IQR 169-236 mg/dL) in women and 185 mg/dL (IQR 152-216 mg/dL) in men. After adjusting for demographic, socioeconomic, lifestyle and dietary factors, participants who reported using palm oil as the major cooking oil had serum total cholesterol higher by 15 mg/dL (95% CI 1 to 29 mg/dL) compared with those who reported using sunflower oil. Consumption of one or more servings of meat per day (p for trend=0.017) and less than five servings of fruits and vegetables per day (p for trend=0.024) were also associated with higher serum total cholesterol. A combination of using palm oil for cooking, eating more than one serving of meat per day and fewer than five servings of fruits and vegetables per day, was associated with 46 mg/dL (95% CI 16 to 76 mg/dL) higher serum total cholesterol. CONCLUSIONS: Using palm oil for cooking was associated with higher serum total cholesterol levels in this peri-urban population in Dar es Salaam. Reduction of saturated fat content of edible oil may be considered as a population-based strategy for primary prevention of cardiovascular diseases.

Clinical, Sonographic, and Pathological Characteristics of RAS-Positive Versus BRAF-Positive Thyroid Carcinoma.

CONTEXT: Mutations in the BRAF and RAS oncogenes are responsible for most well-differentiated thyroid cancer. Yet, our clinical understanding of how BRAF-positive and RAS-positive thyroid cancers differ is incomplete. OBJECTIVE: We correlated clinical, radiographic, and pathological findings from patients with thyroid cancer harboring a BRAF or RAS mutation. DESIGN: Prospective cohort study. SETTING: Academic, tertiary care hospital. PATIENTS: A total of 101 consecutive patients with well-differentiated thyroid cancer. MAIN OUTCOME MEASURE: We compared the clinical, sonographic, and pathological characteristics of patients with BRAF-positive cancer to those with RAS-positive cancer. RESULTS: Of 101 patients harboring these mutations, 71 were BRAF-positive, whereas 30 were RAS-positive. Upon sonographic evaluation, RAS-positive nodules were significantly larger (P = .04), although BRAF-positive nodules were more likely to harbor concerning sonographic characteristics (hypoechogenicity [P < .001]; irregular margins [P = .04]). Cytologically, 70% of BRAF-positive nodules were classified positive for PTC, whereas 87% of RAS-positive nodules were indeterminate (P < .001). Histologically, 96% of RAS-positive PTC malignancies were follicular variants of PTC, whereas 70% of BRAF-positive malignancies were classical variants of PTC. BRAF-positive malignancies were more likely to demonstrate extrathyroidal extension (P = .003), lymphovascular invasion (P = .02), and lymph node metastasis (P < .001). CONCLUSIONS: BRAF-positive malignant nodules most often demonstrate worrisome sonographic features and are frequently associated with positive or suspicious Bethesda cytology. In contrast, RAS-positive malignancy most often demonstrates indolent sonographic features and more commonly associates with lower risk, "indeterminate" cytology. Because BRAF and RAS mutations are the most common molecular perturbations associated with well-differentiated thyroid cancer, these findings may assist with improved preoperative risk assessment by suggesting the likely molecular profile of a thyroid cancer, even when postsurgical molecular analysis is unavailable.

Stroke Is the Leading Cause of Death in Rural Gadchiroli, India: A Prospective Community-Based Study.

BACKGROUND AND PURPOSE: Stroke is an important cause of death and disability worldwide. However, information on stroke deaths in rural India is scarce. To measure the mortality burden of stroke, we conducted a community-based study in a rural area of Gadchiroli, one of the most backward districts of India. METHODS: We prospectively collected information on all deaths from April 2011 to March 2013 and assigned causes of death using a well-validated verbal autopsy tool in a rural population of 94 154 individuals residing in 86 villages. Two trained physicians independently assigned the cause of death, and the disagreements were resolved by a third physician. RESULTS: Of 1599 deaths during the study period, 229 (14.3%) deaths were caused by stroke. Stroke was the most frequent cause of death. For those who died because of stroke, the mean age was 67.47±11.8 years and 48.47% were women. Crude stroke mortality rate was 121.6 (95% confidence interval, 106.4-138.4), and age-standardized stroke mortality rate was 191.9 (95% confidence interval, 165.8-221.1) per 100,000 population. Of total stroke deaths, 87.3% stroke deaths occurred at home and 46.3% occurred within the first month from the onset of symptoms. CONCLUSIONS: Stroke is the leading cause of death and accounted for 1 in 7 deaths in this rural community in Gadchiroli. There was high early mortality, and the mortality rate because of stroke was higher than that reported from previous studies from India. Stroke is emerging as a public health priority in rural India.