Probability Cut-point Research Articles

Clinical and health care utilization variables can predict 90-day hospital re-admission in adults with Crohn’s disease for point of care risk evaluation

BackgroundHospital re-admission for persons with Crohn’s disease (CD) is a significant contributor to morbidity and healthcare costs. We derived prediction models of risk of 90-day re-hospitalization among persons with CD that could be applied at hospital discharge to target outpatient interventions mitigating this risk.MethodsWe performed a retrospective study in persons with CD admitted between 2009 and 2016 for an acute CD-related indication. Demographic, clinical, and health services predictor variables were ascertained through chart review and linkage to administrative health databases. We derived and internally validated a multivariable logistic regression model of 90-day CD-related re-hospitalization. We selected the optimal probability cut-point to maximize Youden’s index.ResultsThere were 524 CD hospitalizations and 57 (10.9%) CD re-hospitalizations within 90 days of discharge. Our final model included hospitalization within the prior year (adjusted odds ratio [aOR] 3.27, 95% confidence interval [CI] 1.76–6.08), gastroenterologist consultation within the prior year (aOR 0.185, 95% CI 0.0950–0.360), intra-abdominal surgery during index hospitalization (aOR 0.216, 95% CI 0.0500–0.934), and new diagnosis of CD during index hospitalization (aOR 0.327, 95% CI 0.0950–1.13). The model demonstrated good discrimination (optimism-corrected c-statistic value 0.726) and excellent calibration (Hosmer-Lemeshow goodness-of-fit p-value 0.990). The optimal model probability cut point allowed for a sensitivity of 71.9% and specificity of 70.9% for identifying 90-day re-hospitalization, at a false positivity rate of 29.1% and false negativity rate of 28.1%.ConclusionsDemographic, clinical, and health services variables can help discriminate persons with CD at risk of early re-hospitalization, which could permit targeted post-discharge intervention.

Estimating Dementia Prevalence in Rural South Africa with Remote Consensus Diagnoses and Algorithmic Modeling

AbstractBackgroundThere is limited population‐based data on dementia epidemiology from sub‐Saharan Africa, precluding deeper understanding of global dementia burden. The Health and Aging in Africa Study (HAALSI) is designed to address this evidence gap through collection of high‐quality longitudinal data on cognition, health, and everyday function in older Black South Africans. Consensus‐based diagnoses have become a gold standard for ascertaining dementia status in research settings, however, this approach is resource‐intensive and difficult to implement at scale. Here, we utilize algorithmic methods to estimate dementia prevalence and risk factors in a population‐representative cohort in rural South Africa.MethodWe utilize data from 4,176 respondents who completed assessments of health, demographics, socioeconomic conditions, cognition, and activities of daily living (ADLs) as part of HAALSI Wave 2 (2018‐2019). An enriched sub‐cohort (n = 635) later completed clinical and neuropsychological assessments and received diagnostic classification of normal cognition, mild cognitive impairment, or dementia by an expert panel. Logistic regression was used to predict dementia status in the enriched sub‐cohort using predictor variables drawn from the parent HAALSI wave. Probability cut points were selected to achieve overall prediction accuracy of at least 80%, prioritizing specificity over sensitivity. Algorithm performance was evaluated across education‐level and age subgroups. Coefficients from best‐performing algorithms were applied to the full cohort to obtain dementia probability scores and calculate dementia prevalence.ResultWhen the enriched sub‐sample was reweighted to reflect the full HAALSI population, the estimated prevalence of dementia was 18%. Three models of increasing complexity were tested and showed good discrimination between dementia and non‐dementia (ROC = 0.76‐0.82). Models were generally more sensitive but less accurate in older and less educated respondents. Algorithmic estimates of dementia prevalence in the HAALSI population ranged from 15‐19%. Higher risk for dementia was associated with older age, lower education and wealth, hypertension, living alone, poor self‐rated health, widowhood, and immigration status.ConclusionThis is one of the first studies to utilize algorithmic methods to describe dementia prevalence in South Africa, producing higher estimates than most previous reports. These efforts may provide a foundation to expand understanding of dementia epidemiology in a region of the world currently experience rapid population aging.

Binary search of the optimal cut-point value in ROC analysis using the F1 score

This paper introduces a binary search algorithm for determining the optimal probability cut-point value (C) of binary classifiers. Cut-points are operating points on the receiver operating characteristic curve that divide positive and negative predictions. Compared to the traditional exhaustive search for optimal C value, the proposed method offers execution time efficiency (O(log2(k))) and a small cut-point error of 1/2 n after k steps of binary search. Traditionally, the optimal C value is determined by stepping through all possible C values. This search is uninformed because there is no indication of the search direction. To address this issue, we derive the expectation of the F-Measure (aka F1 score); and use it to guide the search process. Specifically, by comparing the F-Measure at the current cut-point with the F-Measure at expected cut-point, we can use the information to adjust C dynamically towards the optimal cut-point, resulting in optimal model performance. Our results on two classifiers trained from disease classification datasets suggest that the algorithm is robust and efficient, as compared to the traditional methods.

Predicting Limited Survival Following Inguinal Lymph Node Dissection in Penile Cancer: Should We Revisit the Goals of Care?

Patients with advanced penile squamous cell cancer have a poor prognosis and can benefit from early palliative care consultation. We built a model to identify those patients most likely to benefit. Patients with penile squamous cell cancer undergoing inguinal lymph node dissection were identified from the National Cancer Database (NCDB) and a multi-institutional international dataset (INT). A multivariable Cox proportional hazards model for overall survival (OS) was developed using the NCDB and applied to the INT dataset. Parameters were used to make receiver operating characteristic (ROC) curves. ROC-related criteria were optimized to identify a predictive probability cut point and dichotomize patients from INT into risk groups for limited OS of <6 and <12 months. NCDB had 860 deaths; 105 (5%) at 6 months and 296 (15%) at 12 months. INT had 257 deaths; 56 (8%) at 6 months and 124 (18%) at 12 months. Limited OS was associated with older age, greater T and N stage, and fewer lymph nodes removed. Optimized ROC criteria using the OS <6 months curve best dichotomized INT patients into high-risk group with median OS of 24 months (95% CI 18-34) and low-risk group with median OS of 174 months (95% CI 120-NE). We developed a simple model that could be used as a screening tool for early palliative care referral.

A166 RISK STRATIFICATION OF EARLY RE-HOSPITALIZATION IN PERSONS WITH INFLAMMATORY BOWEL DISEASES USING MULTIVARIABLE MODELS

Abstract Background Hospitalization for persons with inflammatory bowel disease (IBD), including Crohn’s disease (CD) and ulcerative colitis (UC), is a significant contributor to morbidity and health care costs in Canada. Recognition of individuals at high risk of re-hospitalization could help inform targeted outpatient interventions that mitigate this risk. Purpose The aim of our study is to derive prediction models of risk of early (90-day) re-hospitalization among persons with IBD. Method We conducted a retrospective cohort study of all adult persons with IBD admitted to The Ottawa Hospital, Canada, for an acute IBD-related indication between April 2009 - March 2016. Demographic, clinical, and health services variables were obtained through chart review. Persons were linked to population-based health administrative datasets to identify historical and future IBD-related hospitalizations across the greater Ottawa region. Multivariable logistic regression models of 90-day re-hospitalization in persons with CD and UC were derived, and candidate predictors that demonstrated an independent association with the outcome at a p-value of 0.1 were retained. Bootstrap internal validation (200 iterations) was performed on the final models. Model performance and calibration were evaluated using the optimism-corrected c-statistic value and Hosmer-Lemeshow goodness of fit test, respectively. Adjusted odds ratios are reported with 95% confidence intervals (CI). Optimal probability cut points for re-hospitalization were selected to optimize sensitivity, specificity, and the J (Youden’s) index. Result(s) There were 524 CD and 248 UC hospitalizations during the study period. Of these, 57 (10.9%) CD and 27 (10.9%) UC hospitalizations were associated with re-hospitalization within 90 days of discharge. Forty-two candidate predictors were tested among CD hospitalizations, and 35 were tested among UC hospitalizations. Four variables were retained in each of the final models. Model performance and calibration for each variable are described in Table 1. The optimal range of probability cut points allowed for a sensitivity/positive predictive value (PPV)/false positive rate (FPR) of 0.72/0.23/0.29 (maximum J-index of 0.43) in the model for CD, and 0.78/0.33/0.19 (maximum J-index of 0.59) in the model for UC, respectively. Image Conclusion(s) Demographic, clinical, and health services variables at the time of discharge have the potential to help identify persons with IBD at risk of early re-hospitalization, thereby permitting targeted outpatient intervention. Application of the models to our reference cohorts would earmark 1/3 or less of patients for early post-discharge intervention, with the potential to benefit more than 70% of patients destined for early re-hospitalization. Although the PPVs of our models were low, the models incorrectly predicted early re-hospitalization in less than 30% of patients. We are in process of externally validating these models in other jurisdictions across Ontario to test their generalizability. Please acknowledge all funding agencies by checking the applicable boxes below None Disclosure of Interest None Declared

Acid-base imbalances and the association of blood-gas variables, electrolytes, and biochemical analytes with outcome in hospitalized calves undergoing abdominal surgery.

Surgical abdominal emergencies in calves are associated with a guarded prognosis and have the potential for complex metabolic derangements including acid-base imbalances. To perform a comprehensive analysis of acid-base status and to assess the prognostic relevance of preoperative clinicopathologic variables in calves undergoing abdominal surgery. Hospital-based study samples of 535 (dataset 1; DS1) and 83 calves (dataset 2; DS2). Retrospective (DS1) and prospective (DS2) case series. In DS1, acidemia (pH <7.33) was present in 49.9%, whereas alkalemia (pH >7.37) was present in 30.7% of calves. Plasma L-lactate, chloride, and serum inorganic phosphorus concentration accounted for 51.9%, 11.6% and 9.4% of the variation of venous blood pH, respectively. Classification tree analysis indicated that a negative outcome (death or euthanasia during hospitalization) was associated with venous pO2 ≤33.6 mm Hg, anion gap >18.3 and >22.9mEq/L, serum albumin concentration ≤36.5 and ≤29.4g/L, serum urea concentration >4.4 mmol/L, and plasma ionized calcium concentration ≤1.26 mmol/L. The area under the receiver operating characteristic curve of this model was 0.85 (95% CI: 0.82-0.89, P < .001) and the resulting sensitivity and specificity for the prediction of nonsurvival at the optimal probability cut-point of 0.5 was 89.8% and 65.7%, respectively. In DS2 the model had a similar sensitivity and specificity of 90.5% and 70%, respectively. Clinicopathologic imbalances and associated changes of acid-base status are common in calves with surgical abdominal emergencies and have clinical utility for the prediction of a negative postoperative outcome.

Risk stratification by sex and menopausal status in the multivariable apnea prediction index.

To determine the sensitivity of the Multivariable Apnea Prediction (MAP) index for obstructive sleep apnea (OSA) in pre- and post-menopausal women with the goal of developing a tailored scoring classification approach. Data from two studies (N = 386); the diabetes sleep treatment trial (N = 236) and EMPOWER (N = 150) were used to assess the sensitivity and specificity of the MAP index by comparing men (n = 129) to women (n = 257), and premenopausal (n = 100) to post-menopausal women (n = 136). We evaluated participants at two cut points, apnea-hypopnea index (AHI) values of ≥ 5 and ≥ 10, using 0.5 as a predicted probability cut point to establish baseline sensitivity and specificity. Contingency tables and receiver operating characteristic (ROC) analysis were conducted to evaluate the accuracy of the MAP index in predicting OSA in men versus women, and in pre-versus post-menopausal women. To select optimal predicted probabilities for classification by sex and menopausal status, Youden's J statistic was generated from ROC coordinates. The MAP index was more sensitive to women in the AHI ≥ 5 group (76%) compared to AHI ≥ 10 group (30%). Among post-menopausal women with AHI ≥ 5, sensitivity was similar to men (98%), but less than men when AHI ≥ 10 (32%). Suggested probability cut points for women with an AHI ≥ 10 are 0.24 overall; 0.15 for premenopausal, and 0.38 for postmenopausal women. Because women's risk for OSA (AHI ≥ 10) was underestimated by the MAP index, we suggest the use of tailored cut points based on sex and menopausal status or assessing for OSA risk with an AHI of ≥ 5.

Simple clinical and laboratory predictors to improve empirical treatment strategies in areas of high scrub typhus and dengue endemicity, central Vietnam.

BackgroundDengue fever is highly endemic in Vietnam, but scrub typhus—although recognized as an endemic disease—remains underappreciated. These diseases together are likely to account for more than half of the acute undifferentiated fever burden in Vietnam. Scrub typhus (ST) is a bacterial disease requiring antimicrobial treatment, while dengue fever (DF) is of viral etiology and does not. The access to adequate diagnostics and the current understanding of empirical treatment strategies for both illnesses remain limited. In this study we aimed to contribute to the clinical decision process in the management of these two important etiologies of febrile illness in Vietnam.MethodsUsing retrospective data from 221 PCR-confirmed scrub typhus cases and 387 NS1 protein positive dengue fever patients admitted to five hospitals in Khanh Hoa province (central Vietnam), we defined predictive characteristics for both diseases that support simple clinical decision making with potential to inform decision algorithms in future. We developed models to discriminate scrub typhus from dengue fever using multivariable logistic regression (M-LR) and classification and regression trees (CART). Regression trees were developed for the entire data set initially and pruned, based on cross-validation. Regression models were developed in a training data set involving 60% of the total sample and validated in the complementary subsample. Probability cut points for the distinction between scrub typhus and dengue fever were chosen to maximise the sum of sensitivity and specificity.ResultsUsing M-LR, following seven predictors were identified, that reliably differentiate ST from DF; eschar, regional lymphadenopathy, an occupation in nature, increased days of fever on admission, increased neutrophil count, decreased ratio of neutrophils/lymphocytes, and age over 40. Sensitivity and specificity of predictions based on these seven factors reached 93.7% and 99.5%, respectively. When excluding the “eschar” variable, the values dropped to 76.3% and 92.3%, respectively.The CART model generated one further variable; increased days of fever on admission, when eschar was included, the sensitivity and specificity was 95% and 96.9%, respectively. The model without eschar involved the following six variables; regional lymphadenopathy, increased days of fever on admission, increased neutrophil count, increased lymphocyte count, platelet count ≥ 47 G/L and age over 28 years as predictors of ST and provided a sensitivity of 77.4% and a specificity of 90.7%.ConclusionsThe generated algorithms contribute to differentiating scrub typhus from dengue fever using basic clinical and laboratory parameters, supporting clinical decision making in areas where dengue and scrub typhus are co-endemic in Vietnam.

Additive prognostic value of preoperative plasma glucose concentrations in calves undergoing abdominal surgery

Surgical abdominal emergencies in calves are associated with a guarded prognosis, especially if neonates are affected. Because hypoglycemia has been associated with sepsis and endotoxemia, this study aimed to assess the prognostic relevance of preoperative plasma glucose concentrations (GLUC) in calves requiring surgery for an acute abdominal disorder. For this purpose, data from retrospective and prospective case series were analyzed, consisting of 586 and 83 hospitalized calves, respectively. The outcomes of calves were evaluated until hospital discharge (both study populations) and for 3 mo following discharge by a phone call to the farmer (prospective study population). For the retrospective study population, the overall survival rate was 31.2%. Calves with a negative outcome (NO) had significantly lower median GLUC (4.3 mmol/L) than calves with a positive outcome (PO; 5.0 mmol/L). The survival rates of calves with GLUC <2.4 mmol/L and 2.4 to 3.1 mmol/L were 3.6 and 8.3%, respectively. The inclusion of GLUC improved a previous prognostic model based on plasma l-lactate concentration and age. The resulting analyses indicated that NO was associated with low age (<7 d), hyper-l-lactatemia (>8.84 mmol/L), and GLUC <4.4 mmol/L (age 7–20 d) and <3.3 mmol/L (age ≥21 d), respectively. The area under the receiver operating characteristic curve of this model was 0.79 (95% confidence interval: 0.76–0.83) and the resulting sensitivity and specificity for NO at the optimal probability cut-point of 0.69 were 66.7 and 85.8%, respectively. For the prospective study population, the established model had sensitivity and specificity for predicting NO after 3 mo (proportion 24%) of 61.9 and 85%, respectively. In both study populations, hypoglycemia was significantly associated with intraoperative evidence of a septic process within the abdominal cavity. The present analyses show that hypoglycemia was highly indicative of a poor prognosis and serious intraoperative findings such as peritonitis. Determination of GLUC should therefore be part of the diagnostic work-up in calves suffering from an acute abdominal emergency.

PD53-10 PREDICTING LIMITED SURVIVAL FOR PATIENTS WITH PENILE CANCER UNDERGOING INGUINAL LYMPH NODE DISSECTION

PD53-10 PREDICTING LIMITED SURVIVAL FOR PATIENTS WITH PENILE CANCER UNDERGOING INGUINAL LYMPH NODE DISSECTION

A simple assessment of lung nodule location for reduction in unnecessary invasive procedures.

BackgroundCT screening for lung cancer results in a significant mortality reduction but is complicated by invasive procedures performed for evaluation of the many detected benign nodules. The purpose of this study was to evaluate measures of nodule location within the lung as predictors of malignancy.MethodsWe analyzed images and data from 3,483 participants in the National Lung Screening Trial (NLST). All nodules (4–20 mm) were characterized by 3D geospatial location using a Cartesian coordinate system and evaluated in logistic regression analysis. Model development and probability cutpoint selection was performed in the NLST testing set. The Geospatial test was then validated in the NLST testing set, and subsequently replicated in a new cohort of 147 participants from The Detection of Early Lung Cancer Among Military Personnel (DECAMP) Consortium.ResultsThe Geospatial Test, consisting of the superior-inferior distance (Z distance), nodule diameter, and radial distance (carina to nodule) performed well in both the NLST validation set (AUC 0.85) and the DECAMP replication cohort (AUC 0.75). A negative Geospatial Test resulted in a less than 2% risk of cancer across all nodule diameters. The Geospatial Test correctly reclassified 19.7% of indeterminate nodules with a diameter over 6mm as benign, while only incorrectly classifying 1% of cancerous nodules as benign. In contrast, the parsimonious Brock Model applied to the same group of nodules correctly reclassified 64.5% of indeterminate nodules as benign but resulted in misclassification of a cancer as benign in 18.2% of the cases. Applying the Geospatial test would result in reducing invasive procedures performed for benign lesions by 11.3% with a low rate of misclassification (1.3%). In contrast, the Brock model applied to the same group of patients results in decreasing invasive procedures for benign lesion by 39.0% but misclassifying 21.1% of cancers as benign.ConclusionsUtilizing information about geospatial location within the lung improves risk assessment for indeterminate lung nodules and may reduce unnecessary procedures.Trial RegistrationNCT00047385, NCT01785342.

Development and validation of a multivariable predictive model for EGFR gene mutation status in patients with lung adenocarcinoma.

Detection of epidermal growth factor receptor (EGFR) is one real dilemma owing to the non-sufficient tissue for testing EGFR mutations in lung adenocarcinoma. A model for predicting EGFR mutations would be helpful for clinical decisions in those patients. A retrospective cohort of 1,196 patients diagnosed with lung adenocarcinoma was investigated between December 1, 2017, and December 31, 2019, in Renji Hospital, Shanghai, China. All patients were tested for EGFR mutations (amplification refractory mutation system, n=1,144; next-generation sequencing, n=52). Of 1,196 patients with lung adenocarcinoma, 944 met the inclusion criteria. A nomogram model was developed based on 567 patients and validated in 377 patients. Variables associated with EGFR mutations were age, sex, smoking history, lepidic predominant subtype, solid predominant subtype, mucinous adenocarcinoma, Ki67 expression, lobulation, solid texture in radiology, and pleural retraction. The nomogram based on the model performed well in the development group (c-index 0.789, 95% CI: 0.751-0.827), and the validation group (c-index 0.809, 95% CI: 0.771-0.847). At the probability cut-point of 0.7, the diagnostic efficiency was 82.7% in patients with NGS liquid biopsy. Decision curve analysis further confirmed the clinical usefulness of the nomogram, which showed that predicting the EGFR mutations probability applying this nomogram would be better than having all patients or none patients use this nomogram. A high probability group (>0.7) by nomogram model may suggest a high possibility of EGFR mutation, if tissue is limited, NGS-based ctDNA with liquid biopsy could be implemented effectively.

Predicting Care Homes with Limited Residential Information

IntroductionResearch involving care homes is often difficult due to a lack of data and ethical issues. Wales (United Kingdom) contains approximately 1.3million residences, of these 717 are officially recorded as care homes for older people.&#x0D; Objectives and ApproachOur objective was to develop a predictive methodology for identifying care homes in administrative data.&#x0D; We used two data sources within the Secure Anonymised Information Linkage Databank to conduct our study. The Welsh Demographic Service Dataset (WDSD) contains all residences in Wales and demographic details of their occupants. An anonymised dataset of deterministically matched care home addresses was used to determine which of the residences in the WDSD were care homes.&#x0D; We used details in the WDSD to determine the average age of the occupants, the number of people who moved into the residence in a year, and the number of people who died in a year. We were interested in care homes for older people and restricted all the residences in the WDSD to only those with an average age of 50+ years. We applied logistic regression to determine a probabilistic match for care homes based on the above characteristics. We determined an optimal cut-point for the probability of a residence being a care home based on the sensitivity and specificity.&#x0D; ResultsRestricting the WDSD to have an average age of occupants of 50+ created a dataset of 3,939 residences, containing 562 care homes. After applying the logistic model to predict the care homes, we found an optimal probability cut-point which resulted in 548 true positives, 105 false positives, 14 false negatives, and 3,272 true negatives.&#x0D; ImplicationsIdentification of care homes in an anonymised databank using only demographic data allows research into healthcare pathways for this hard to reach and under-researched population.

Can Machine Learning Methods Produce Accurate and Easy-to-Use Preoperative Prediction Models of One-Year Improvements in Pain and Functioning After Knee Arthroplasty?

BackgroundApproximately 15%-20% of total knee arthroplasty (TKA) patients do not experience clinically meaningful improvements. We sought to compare the accuracy and parsimony of several machine learning strategies for developing predictive models of failing to experience minimal clinically important differences in patient-reported outcome measures (PROMs) 1 year after TKA. MethodsPatients (N = 587) in 3 large Veteran Health Administration facilities completed PROMs before and 1 year after TKA (92% follow-up). Preoperative PROMs and electronic health record data were used to develop and validate models to predict failing to experience at least a minimal clinically important difference in Knee Injury and Osteoarthritis Outcome Score (KOOS) Total, KOOS JR, and KOOS subscales (Pain, Symptoms, Activities of Daily Living, Quality of Life, and recreation). Several machine learning strategies were used for model development. Ten-fold cross-validation and bootstrapping were used to produce measures of overall accuracy (C-statistic, Brier Score). The sensitivity and specificity of various predicted probability cut-points were examined. ResultsThe most accurate models produced were for the Activities of Daily Living, Pain, Symptoms, and Quality of Life subscales of the KOOS (C-statistics 0.76, 0.72, 0.72, and 0.71, respectively). Strategies varied substantially in terms of the numbers of inputs required to achieve similar accuracy, with none being superior for all outcomes. ConclusionModels produced in this project provide estimates of patient-specific improvements in major outcomes 1 year after TKA. Integrating these models into clinical decision support, informed consent and shared decision making could improve patient selection, education, and satisfaction. Level of EvidenceLevel III, diagnostic study.

Development and validation of a predictive model for estimating EGFR mutation probabilities in patients with non-squamous non-small cell lung cancer in New Zealand

BackgroundTargeted treatment with Epidermal Growth Factor Receptor (EGFR) tyrosine kinase inhibitors (TKIs) is superior to systemic chemotherapy in non-small cell lung cancer (NSCLC) patients with EGFR gene mutations. Detection of EGFR mutations is a challenge in many patients due to the lack of suitable tumour specimens for molecular testing or for other reasons. EGFR mutations are more common in female, Asian and never smoking NSCLC patients.MethodsPatients were from a population-based retrospective cohort of 3556 patients diagnosed with non-squamous non-small cell lung cancer in northern New Zealand between 1 Feb 2010 and 31 July 2017. A total of 1694 patients were tested for EGFR mutations, of which information on 1665 patients was available for model development and validation. A multivariable logistic regression model was developed based on 1176 tested patients, and validated in 489 tested patients. Among 1862 patients not tested for EGFR mutations, 129 patients were treated with EGFR-TKIs. Their EGFR mutation probabilities were calculated using the model, and their duration of benefit and overall survival from the start of EGFR-TKI were compared among the three predicted probability groups: < 0.2, 0.2–0.6, and > 0.6.ResultsThe model has three predictors: sex, ethnicity and smoking status, and is presented as a nomogram to calculate EGFR mutation probabilities. The model performed well in the validation group (AUC = 0.75). The probability cut-point of 0.2 corresponds 68% sensitivity and 78% specificity. The model predictions were related to outcome in a group of TKI-treated patients with no biopsy testing available (n = 129); in subgroups with predicted probabilities of < 0.2, 0.2–0.6, and > 0.6, median overall survival times from starting EGFR-TKI were 4.0, 5.5 and 18.3 months (p = 0.02); and median times remaining on EGFR-TKI treatment were 2.0, 4.2, and 14.0 months, respectively (p < 0.001).ConclusionOur model may assist clinical decision making for patients in whom tissue-based mutation testing is difficult or as a supplement to mutation testing.

Development and Evaluation of a Prediction Model for Ascertaining Rheumatic Heart Disease Status in Administrative Data.

BackgroundPrevious research has raised substantial concerns regarding the validity of the International Statistical Classification of Diseases and Related Health Problems (ICD) codes (ICD-10 I05–I09) for rheumatic heart disease (RHD) due to likely misclassification of non-rheumatic valvular disease (non-rheumatic VHD) as RHD. There is currently no validated, quantitative approach for reliable case ascertainment of RHD in administrative hospital data.MethodsA comprehensive dataset of validated Australian RHD cases was compiled and linked to inpatient hospital records with an RHD ICD code (2000–2018, n=7555). A prediction model was developed based on a generalized linear mixed model structure considering an extensive range of demographic and clinical variables. It was validated internally using randomly selected cross-validation samples and externally. Conditional optimal probability cutpoints were calculated, maximising discrimination separately for high-risk versus low-risk populations.ResultsThe proposed model reduced the false-positive rate (FPR) from acute rheumatic fever (ARF) cases misclassified as RHD from 0.59 to 0.27; similarly for non-rheumatic VHD from 0.77 to 0.22. Overall, the model achieved strong discriminant capacity (AUC: 0.93) and maintained a similar robust performance during external validation (AUC: 0.88). It can also be used when only basic demographic and diagnosis data are available.ConclusionThis paper is the first to show that not only misclassification of non-rheumatic VHD but also of ARF as RHD yields substantial FPRs. Both sources of bias can be successfully addressed with the proposed model which provides an effective solution for reliable RHD case ascertainment from hospital data for epidemiological disease monitoring and policy evaluation.

Early detection of chronic kidney disease in low-income and middle-income countries: development and validation of a point-of-care screening strategy for India

IntroductionAlthough deaths due to chronic kidney disease (CKD) have doubled over the past two decades, few data exist to inform screening strategies for early detection of CKD in low-income and...

Prognostic value of preoperative plasma l-lactate concentrations in calves with acute abdominal emergencies

Acute abdominal emergencies in calves due to abomasal disorders, gastrointestinal ileus, or peritonitis are characterized by a rapid disease progression and usually require immediate surgical intervention. Those conditions are associated with a guarded prognosis, and the aim of the present study was to assess the prognostic relevance of preoperatively measured plasma l-lactate concentrations (l-LAC) in a large study population of calves with a broad spectrum of acute abdominal emergencies. For the purpose of this study, the medical records of 587 calves admitted to a veterinary teaching hospital over a 10-yr period were analyzed retrospectively. Plasma l-LAC was measured as part of a routinely performed biochemistry panel before initiation of surgical intervention. Hyper-l-lactatemia (plasma l-LAC >2.2 mmol/L) was evident in 75% of calves, and the overall survival rate until hospital discharge was 31%. Calves with a negative outcome were younger (median: 3.4 vs. 6 wk) and had higher plasma l-LAC (median: 4.96 vs. 3.09 mmol/L) than calves with a positive outcome. At the individual diagnosis level, l-LAC was associated with mortality in calves with a diagnosis of mesenteric torsion, right-sided dilated abomasum, small intestinal volvulus, or paralytic ileus, but not in calves suffering from peritonitis, malformations, abomasal volvulus, bloat, or small intestinal intussusceptions. Considering the whole study population, the area under the receiver operating characteristic (ROC) curve for plasma l-LAC was 0.66 [95% confidence interval (CI): 0.61-0.70]. A classification tree analysis indicated that l-LAC >8.84 mmol/L and age categories of <3 wk and <1 wk were independent predictors of mortality. The area under the ROC curve of this model was 0.75 (95% CI: 0.71-0.79) and the resulting sensitivity and specificity for the prediction of nonsurvival at the optimal probability cut-point of 0.62 were 67.7 and 76.6%, respectively. In conclusion, hyper-l-lactatemia is common in calves suffering from acute abdominal emergencies. Markedly increased plasma l-LAC is associated with an increased mortality risk, but it is not possible to reliably predict the outcome of affected calves based on a single, preoperative measurement. However, a clinically important finding of this study was that the ability to predict a negative outcome is improved when the age of the calf is considered in addition to plasma l-LAC.

Validation of an Electronic Medical Record-Based Algorithm for Identifying Posttraumatic Stress Disorder in U.S. Veterans.

We developed an algorithm for identifying U.S. veterans with a history of posttraumatic stress disorder (PTSD), using the Department of Veterans Affairs (VA) electronic medical record (EMR) system. This work was motivated by the need to create a valid EMR-based phenotype to identify thousands of cases and controls for a genome-wide association study of PTSD in veterans. We used manual chart review (n = 500) as the gold standard. For both the algorithm and chart review, three classifications were possible: likely PTSD, possible PTSD, and likely not PTSD. We used Lasso regression with cross-validation to select statistically significant predictors of PTSD from the EMR and then generate a predicted probability score of being a PTSD case for every participant in the study population (range: 0-1.00). Comparing the performance of our probabilistic approach (Lasso algorithm) to a rule-based approach (International Classification of Diseases [ICD] algorithm), the Lasso algorithm showed modestly higher overall percent agreement with chart review than the ICD algorithm (80% vs. 75%), higher sensitivity (0.95 vs. 0.84), and higher accuracy (AUC = 0.95 vs. 0.90). We applied a 0.7 probability cut-point to the Lasso results to determine final PTSD case-control status for the VA population. The final algorithm had a 0.99 sensitivity, 0.99 specificity, 0.95 positive predictive value, and 1.00 negative predictive value for PTSD classification (grouping possible PTSD and likely not PTSD) as determined by chart review. This algorithm may be useful for other research and quality improvement endeavors within the VA.

Development and validation of a claims-based approach to proxy ECOG performance status across ten tumor groups

To develop a claims-based prediction model of poor performance status (PS) in commercially insured and Medicare supplemental beneficiaries with cancer. Retrospective analysis was conducted of electronic medical records (EMR) from community oncology practices linked to MarketScan claims. Multivariable logistic regression predicted PS scores from the EMR using claims-based diagnostic and procedure codes. The study included 8442patients diagnosed with cancer from 2007 to 2015. Overall, 8.1% of patients had poor EMR-based PS. Bootstrapping results from the final model showed sensitivity and specificity of approximately 75% with a predicted probability cutpoint=0.078, c-statistic=0.821 and pseudo-R2 =0.25. Patients with poor PS can be identified in claims data. This prediction model enables future studies evaluating cancer treatments and outcomes to account for PS.