Model-building Set Research Articles

Component prediction in combined hepatocellular carcinoma-cholangiocarcinoma: habitat imaging and its biologic underpinnings.

To construct an MRI-based habitat imaging model to help predict component percentage in combined hepatocellular carcinoma-cholangiocarcinoma (cHCC-CCA) preoperatively, and investigate the biologic underpinnings of habitat imaging in cHCC-CCA. The study consisted of one retrospective model-building dataset and one prospective validation dataset from two hospitals. All voxels were assigned into different clusters according to the similarity of enhancement pattern by using K-means clustering method, and each habitat's volume fraction in each lesion was calculated. Least absolute shrinkage and selection operator (LASSO) regression analysis was performed to select optimal predictors, and then to establish an MRI-based habitat imaging model. R-squared was calculated to evaluate performance of the prediction models. Model performance was also verified in the prospective dataset with RNA sequencing data, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was then applied to investigate the biologic underpinnings of habitat imaging. A total of 129 patients were enrolled (mean age, 56.1 ± 10.4, 102 man), among which 104 patients were in the retrospective model-building set, while 25 patients in the prospective validation set. Three habitats, habitat1 (HCC-alike habitat), habitat2 (iCCA-alike habitat), and habitat3 (in-between habitat), were identified. Habitat 1's volume fraction, habitat 3's volume fraction, nonrim APHE, nonperipheral washout, and LI-RADS categorization were selected to develop an HCC percentage prediction model with R-squared of 0.611 in the model-building set and 0.541 in the validation set. Habitat 1's volume fraction was correlated with genes involved in regulation of actin cytoskeleton and Rap1 signaling pathway, which regulate cell migration and tumor metastasis. Preoperative prediction of HCC percentage in patients with cHCC-CCA was achieved using an MRI-based habitat imaging model, which may correlate with signaling pathways regulating cell migration and tumor metastasis.

Structural equation modeling (SEM) of kidney function markers and longitudinal CVD risk assessment.

Lower kidney function is known to enhance cardiovascular disease (CVD) risk. It is unclear which estimated glomerular filtration rate (eGFR) equation best predict an increased CVD risk and if prediction can be improved by integration of multiple kidney function markers. We performed structural equation modeling (SEM) of kidney markers and compared the performance of the resulting pooled indexes with established eGFR equations to predict CVD risk in a 10-year longitudinal population-based design. We split the study sample into a set of participants with only baseline data (n = 647; model-building set) and a set with longitudinal data (n = 670; longitudinal set). In the model-building set, we fitted five SEM models based on serum creatinine or creatinine-based eGFR (eGFRcre), cystatin C or cystatin-based eGFR (eGFRcys), uric acid (UA), and blood urea nitrogen (BUN). In the longitudinal set, 10-year incident CVD risk was defined as a Framingham risk score (FRS)>5% and a pooled cohort equation (PCE)>5%. Predictive performances of the different kidney function indexes were compared using the C-statistic and the DeLong test. In the longitudinal set, a SEM-based estimate of latent kidney function based on eGFRcre, eGFRcys, UA, and BUN showed better prediction performance for both FRS>5% (C-statistic: 0.70; 95% CI: 0.65-0.74) and PCE>5% (C-statistic: 0.75; 95%CI: 0.71-0.79) than other SEM models and different eGFR formulas (DeLong test p-values<3.21×10-6 for FRS>5% and <1.49×10-9 for PCE>5%, respectively). However, the new derived marker could not outperform eGFRcys (DeLong test p-values = 0.88 for FRS>5% and 0.20 for PCE>5%, respectively). SEM is a promising approach to identify latent kidney function signatures. However, for incident CVD risk prediction, eGFRcys could still be preferrable given its simpler derivation.

Applying machine learning to the pharmacokinetic modeling of cyclosporine in adult renal transplant recipients: a multi-method comparison.

Objective: The aim of this study was to identify the important factors affecting cyclosporine (CsA) blood concentration and estimate CsA concentration using seven different machine learning (ML) algorithms. We also assessed the predictability of established ML models and previously built population pharmacokinetic (popPK) model. Finally, the most suitable ML model and popPK model to guide precision dosing were determined. Methods: In total, 3,407 whole-blood trough and peak concentrations of CsA were obtained from 183 patients who underwent initial renal transplantation. These samples were divided into model-building and evaluation sets. The model-building set was analyzed using seven different ML algorithms. The effects of potential covariates were evaluated using the least absolute shrinkage and selection operator algorithms. A separate evaluation set was used to assess the ability of all models to predict CsA blood concentration. R squared (R 2) scores, median prediction error (MDPE), median absolute prediction error (MAPE), and the percentages of PE within 20% (F20) and 30% (F30) were calculated to assess the predictive performance of these models. In addition, previously built popPK model was included for comparison. Results: Sixteen variables were selected as important covariates. Among ML models, the predictive performance of nonlinear-based ML models was superior to that of linear regression (MDPE: 3.27%, MAPE: 34.21%, F20: 30.63%, F30: 45.03%, R 2 score: 0.68). The ML model built with the artificial neural network algorithm was considered the most suitable (MDPE: -0.039%, MAPE: 25.60%, F20: 39.35%, F30: 56.46%, R 2 score: 0.75). Its performance was superior to that of the previously built popPK model (MDPE: 5.26%, MAPE: 29.22%, F20: 33.94%, F30: 51.22%, R 2 score: 0.68). Furthermore, the application of the most suitable model and the popPK model in clinic showed that most dose regimen recommendations were reasonable. Conclusion: The performance of these ML models indicate that a nonlinear relationship for covariates may help to improve model predictability. These results might facilitate the application of ML models in clinic, especially for patients with unstable status or during initial dose optimization.

MO383: Structural Equation Modelling to Improve Kidney Function Assessment in The General Population

Abstract BACKGROUND AND AIMS In order to identify behavioral, environmental, biological or genetic determinants of kidney function, population-based epidemiological studies rely on approximate estimates of the glomerular filtration rate (eGFR) based on measured markers such as serum creatinine, cystatin C, and a few other [1,2]. Since none of these markers is the exclusive reflection of the kidney function, results should always be interpreted with caution to disentangle determinants of kidney function from those that are specific to the marker metabolism. A structural equation modelling (SEM) approach could overcome these limitations allowing to estimate the unobservable kidney function based on a larger number of markers, improving the possibilities of conducting more specific inferential studies [3]. We implemented SEM to identify a latent kidney function trait in a general population sample and validated this derived biomarker against 10-year incidence of kidney disease using an independent dataset. METHOD In the population-based MICROS study (South Tyrol, Italy) [4], we measured serum creatinine (SCr), uric acid (UA), blood urea nitrogen (BUN), cystatin C (CysC), and serum albumin in 1317 study participants. We split the sample into a model-building (n = 647) and a validation set (n = 670). The model-building set was used for SEM analysis and estimation of factor loadings for each biomarker, using the R package ‘lavaan’. Participants in the validation set were submitted a 10-year follow-up interview on kidney disease within the Cooperative Health Research In South Tyrol (CHRIS) study [5]. We fitted logistic regression models to assess the predictivity of the estimated latent kidney function trait versus each individual marker, CKD-EPI eGFR estimates based on serum creatinine and cystatin C (CKD-EPI 2009; CKD-EPI 2021; CKD-EPI Cys; and CKD-EPI CreCys), as well as MDRD-4 and MDRD-6. We estimated odds ratios (ORs) for 10-year kidney disease incidence per standard deviation (SD) change in each marker. RESULTS In the model-building set (mean age 49 years, SD = 19; 58% females; mean eGFRCKD-EPI 2021 = 85.9, SD = 20.9), we identified factor loadings for SCr, UA, BUN, and CysC (Figure 1) achieving acceptable goodness of fit (Chi-square = 14.7; CFI = 0.98; TLI = 0.93; and RMSEA = 0.099). The validation set (baseline mean age = 40 years, SD = 13; 55% females; mean eGFRCKD-EPI 2021 = 100.8, SD = 16.1) had a 10-year kidney disease incidence of 7.8%. For the latent kidney function trait, we observed an OR of 2.1 (95% CI: 1.4–3.2) for incidence per each SD change. This was lower than the ORs (95% CI) associated with the eGFRMDRD4 (2.3, 1.4–4.1) and eGFRMDRD6 (2.5, 1.5–4.2), but larger than the ORs (95% CI) observed for all other markers (SCr: 1.8, 1.3–2.6; UA: 1.2, 0.9–1.8; BUN: 1.3, 0.9–1.8; and CysC: 1.3, 1.0–1.7) or eGFR estimates (eGFRCKD-EPI2009: 1.8, 1.2–2.7; eGFRCKD-EPI2021: 1.7, 1.2–2.5; eGFRCKD-EPI Cys: 1.4, 1.0–1.9; and eGFRCKD-EPI CreCys: 1.8, 1.2–2.6). CONCLUSION A latent kidney function trait derived from SEM performed similarly and often better than conventional eGFR estimates in a 10-year kidney disease incidence prediction. The method warrants validation in independent general population samples with large sample size.

Outcome Prediction in Patients with Severe Traumatic Brain Injury Using Deep Learning from Head CT Scans.

Background After severe traumatic brain injury (sTBI), physicians use long-term prognostication to guide acute clinical care yet struggle to predict outcomes in comatose patients. Purpose To develop and evaluate a prognostic model combining deep learning of head CT scans and clinical information to predict long-term outcomes after sTBI. Materials and Methods This was a retrospective analysis of two prospectively collected databases. The model-building set included 537 patients (mean age, 40 years ± 17 [SD]; 422 men) from one institution from November 2002 to December 2018. Transfer learning and curriculum learning were applied to a convolutional neural network using admission head CT to predict mortality and unfavorable outcomes (Glasgow Outcomes Scale scores 1-3) at 6 months. This was combined with clinical input for a holistic fusion model. The models were evaluated using an independent internal test set and an external cohort of 220 patients with sTBI (mean age, 39 years ± 17; 166 men) from 18 institutions in the Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) study from February 2014 to April 2018. The models were compared with the International Mission on Prognosis and Analysis of Clinical Trials in TBI (IMPACT) model and the predictions of three neurosurgeons. Area under the receiver operating characteristic curve (AUC) was used as the main model performance metric. Results The fusion model had higher AUCs than did the IMPACT model in the prediction of mortality (AUC, 0.92 [95% CI: 0.86, 0.97] vs 0.80 [95% CI: 0.71, 0.88]; P < .001) and unfavorable outcomes (AUC, 0.88 [95% CI: 0.82, 0.94] vs 0.82 [95% CI: 0.75, 0.90]; P = .04) on the internal data set. For external TRACK-TBI testing, there was no evidence of a significant difference in the performance of any models compared with the IMPACT model (AUC, 0.83; 95% CI: 0.77, 0.90) in the prediction of mortality. The Imaging model (AUC, 0.73; 95% CI: 0.66-0.81; P = .02) and the fusion model (AUC, 0.68; 95% CI: 0.60, 0.76; P = .02) underperformed as compared with the IMPACT model (AUC, 0.83; 95% CI: 0.77, 0.89) in the prediction of unfavorable outcomes. The fusion model outperformed the predictions of the neurosurgeons. Conclusion A deep learning model of head CT and clinical information can be used to predict 6-month outcomes after severe traumatic brain injury. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by Haller in this issue.

The predictive value of resting heart rate in identifying undiagnosed diabetes in Korean adults: Korea National Health and Nutrition Examination Survey.

OBJECTIVESThe purpose of this study was (1) to examine whether the addition of resting heart rate (RHR) to the existing undiagnosed diabetes mellitus (UnDM) prediction model would improve predictability, and (2) to develop and validate UnDM prediction models by using only easily assessable variables such as gender, RHR, age, and waist circumference (WC).METHODSKorea National Health and Nutrition Examination Survey (KNHANES) 2010, 2012, 2014, 2016 data were used to develop the model (model building set, n=19,675), while the data from 2011, 2013, 2015, 2017 were used to validate the model (validation set, n=19,917). UnDM was defined as a fasting glucose level ≥126 mg/dL or glycated hemoglobin ≥6.5%; however, doctors have not diagnosed it. Statistical package for the social sciences logistic regression analysis was used to determine the predictors of UnDM.RESULTSRHR, age, and WC were associated with UnDM. When RHR was added to the existing model, sensitivity was reduced (86 vs. 73%), specificity was increased (49 vs. 65%), and a higher Youden index (35 vs. 38) was expressed. When only gender, RHR, age, and WC were used in the model, a sensitivity, specificity, and Youden index of 70%, 67%, and 37, respectively, were observed.CONCLUSIONSAdding RHR to the existing UnDM prediction model improved specificity and the Youden index. Furthermore, when the prediction model only used gender, RHR, age, and WC, the outcomes were not inferior to those of the existing prediction model.

Model for Predicting Breast Cancer Risk in Women With Atypical Hyperplasia

Purpose Women with atypical hyperplasia (AH) on breast biopsy have an aggregate increased risk of breast cancer (BC), but existing risk prediction models do not provide accurate individualized estimates of risk in this subset of high-risk women. Here, we used the Mayo benign breast disease cohort to develop and validate a model of BC risk prediction that is specifically for women with AH, which we have designated as AH-BC. Patients and Methods Retrospective cohorts of women age 18 to 85 years with pathologically confirmed benign AH from Rochester, MN, and Nashville, TN, were used for model development and external validation, respectively. Clinical risk factors and histologic features of the tissue biopsy were selected using L1-penalized Cox proportional hazards regression. Identified features were included in a Fine and Gray regression model to estimate BC risk, with death as a competing risk. Model discrimination and calibration were assessed in the model-building set and an external validation set. Results The model-building set consisted of 699 women with AH, 142 of whom developed BC (median follow-up, 8.1 years), and the external validation set consisted of 461 women with 114 later BC events (median follow-up, 11.4 years). The final AH-BC model included three covariates: age at biopsy, age at biopsy squared, and number of foci of AH. At 10 years, the AH-BC model demonstrated good discrimination (0.63 [95% CI, 0.57 to 0.70]) and calibration (0.87 [95% CI, 0.66 to 1.24]). In the external validation set, the model showed acceptable discrimination (0.59 [95% CI, 0.51 to 0.67]) and calibration (0.91 [95% CI, 0.65 to 1.42]). Conclusion We have created a new model with which to refine BC risk prediction for women with AH. The AH-BC model demonstrates good discrimination and calibration, and it validates in an external data set.

Predicting nonrecovery among whiplash patients in the emergency room and in an insurance company setting.

To construct and validate a prediction instrument for early identification of patients with a high risk of delayed recovery after whiplash injuries (PPS-WAD) in an insurance company setting. Prospective cohort study. On the basis of a historic cohort (n=130) of patients with a whiplash injury identified in an emergency room (ER, model-building set), we used logistic regression to construct an instrument consisting of two demographic variables (i.e. questions of educational level and work status) and the patient-rated physical and mental status during the acute phase to predict self-reported nonrecovery after 6months. We evaluated the instrument's ability to predict nonrecovery in a new cohort (n=204) of patients originating from an insurance company setting (IC, validation set). The prediction instrument had low reproducibility when the setting was changed from the ER cohort to the IC cohort. The overall percentage of correct predictions of nonrecovery in the ER cohort was 78% compared with 62% in the IC cohort. The sensitivity and specificity in relation to nonrecovery were both 78% in the ER cohort. The sensitivity and specificity in the insurance company setting was lower, 67 and 50%. Clinical decision rules need validation before they are used in a new setting. An instrument consisting of four questions with an excellent possibility of identifying patients with a high risk of nonrecovery after a whiplash injury in the emergency room was not as useful in an insurance company setting. The importance and type of the risk factors for not recovering probably differ between the settings, as well as the individuals.

A new model for predicting breast cancer risk in women with atypical hyperplasia.

2 Background: Women with atypical hyperplasia (AH) on breast biopsy have an aggregate increased risk of breast cancer (BC), but accurate personalized risk prediction is desirable to facilitate individual clinical management decisions. Currently used models provide poor BC risk prediction for women with AH. Our goal was to develop and validate an improved risk prediction model for women with AH. Methods: From a cohort of 13,538 women with benign breast disease from 1967-2001, pathology review confirmed 699 with AH. Clinical risk factors and histologic features of the tissue biopsy were recorded, and BC events were ascertained from study questionnaires, tumor registry, and review of medical records. Using a lasso approach, 23 variables were assessed for model inclusion. Lasso-identified features were then fit in a Cox regression model to estimate BC risk. Model discrimination was assessed with C-statistics in the model-building set and in a separate external validation set. Calibration was assessed by comparing observed to predicted breast cancer counts. Results: The model-building set comprised 699 women with 142 BC events (median follow-up 8.1 years), and the external validation set comprised 461 women with 114 BC events (median follow-up 11.4 years). The final model included three covariates: age at biopsy, age squared, and number of foci of AH. Model performance was good, with a C-statistic of 0.622 (SE = 0.027) in the model-building set and 0.594 (SE = 0.029) in the external validation set. The model is well-calibrated, with observed to expected numbers of BCs nearly equal across all post-biopsy follow-up years. Conclusions: We propose a new model for predicting BC risk in women with AH based on age at biopsy and number of foci of atypia. This model provides absolute risk estimates for women with AH, has good discriminatory ability, is well-calibrated, and validates in an external cohort.

Should sleep laboratories have their own predictive formulas for continuous positive airway pressure for patients with obstructive sleep apnea syndrome?

BackgroundSeveral formulas predicting optimal continuous positive airway pressure (CPAP) for obstructive sleep apnea treatment have been developed and diverse parameters selected as predictive factors in different sleep laboratories using different ethnic groups. This study aimed to validate a constructed predictive formula for the study laboratory and to test the hypothesis that sleep laboratories should have their own predictive formulas. MethodsFifty-seven adult subjects with obstructive sleep apnea syndrome (OSAS) were enrolled in the model-building set and underwent two polysomnography (PSG) studies to diagnose OSAS and titrate for optimal CPAP. A predictive formula, derived from anthropometric and polysomnographic variables, was validated together with two other predictive formulas in 30 subjects by comparing the mean predictive CPAP values, rates of successful prediction, and agreements. ResultsRegression analysis showed that apnea–hypopnea index (AHI), SaO2nadir (nadir of arterial oxyhemoglobin saturation by pulse oximetry), and body mass index (BMI) strongly correlated with optimal CPAP. The derived predictive formula for the study laboratory was: CPAPpred (predictive CPAP) = 6.380 + 0.033 × AHI – 0.068 × SaO2nadir + 0.171 × BMI (R2 = 0.335, adjusted R2 = 0.298). In Taiwan, different predictive formulas used by different sleep laboratories with different independent predictors led to similar mean predictive CPAP values to the mean observed optimal CPAP values, rates of successful prediction, and agreements with the observed optimal CPAP. There were significant differences between the mean predictive CPAP values and mean observed optimal CPAP values, lower rates of successful prediction, and negatively skewed 95% confidence interval (CI) when using a predictive formula derived from different ethnic populations. ConclusionA sleep laboratory may not need to have its own predictive formula for determining the optimal effective CPAP but should adopt the one derived from the same ethnicity of OSAS patients as the reference formula.

The common occurrence of epistasis in the determination of human pigmentation and its impact on DNA-based pigmentation phenotype prediction

The role of epistatic effects in the determination of complex traits is often underlined but its significance in the prediction of pigmentation phenotypes has not been evaluated so far. The prediction of pigmentation from genetic data can be useful in forensic science to describe the physical appearance of an unknown offender, victim, or missing person who cannot be identified via conventional DNA profiling. Available forensic DNA prediction systems enable the reliable prediction of several eye and hair colour categories. However, there is still space for improvement. Here we verified the association of 38 candidate DNA polymorphisms from 13 genes and explored the extent to which interactions between them may be involved in human pigmentation and their impact on forensic DNA prediction in particular. The model-building set included 718 Polish samples and the model-verification set included 307 independent Polish samples and additional 72 samples from Japan. In total, 29 significant SNP-SNP interactions were found with 5 of them showing an effect on phenotype prediction. For predicting green eye colour, interactions between HERC2 rs12913832 and OCA2 rs1800407 as well as TYRP1 rs1408799 raised the prediction accuracy expressed by AUC from 0.667 to 0.697 and increased the prediction sensitivity by >3%. Interaction between MC1R 'R' variants and VDR rs731236 increased the sensitivity for light skin by >1% and by almost 3% for dark skin colour prediction. Interactions between VDR rs1544410 and TYR rs1042602 as well as between MC1R 'R' variants and HERC2 rs12913832 provided an increase in red/non-red hair prediction accuracy from an AUC of 0.902-0.930. Our results thus underline epistasis as a common phenomenon in human pigmentation genetics and demonstrate that considering SNP-SNP interactions in forensic DNA phenotyping has little impact on eye, hair and skin colour prediction.

Development of a nomogram for predicting the stone‐free rate after transurethral ureterolithotripsy using semi‐rigid ureteroscope

To develop and to internally validate a novel nomogram for predicting the stone-free rate after transurethral ureterolithotripsy. A total of 412 patients with 534 ureteral stones were treated with transurethral ureterolithotripsy using semi-rigid ureteroscopes. Treatment efficacy was evaluated 3 months after the procedure. Multivariate stepwise logistic regression analysis was used to identify independent predictors of being stone-free in the model-building set. A total of 427 stones (80% of 534) were randomly allocated for identification and statistical analysis to build the model, and the remaining 107 (20%) were used for cross-validation. A nomogram for the stone-free rate was developed based on the final logistic regression model. Stone length, number of stones, stone location and the presence of pyuria were independent factors related to the stone-free rate after transurethral ureterolithotripsy treatment, and these were used to develop a nomogram. In this nomogram, the area under the receiver operating characteristic curve was 0.7432 for the nomogram, 0.5641 for stone size, 0.5908 for the number of stones, 0.6594 for stone location and 0.6076 for pyuria. Validation using 20% of the data also achieved a reasonable predictive accuracy (area under the receiver operating characteristic curve = 0.682). The first nomogram for predicting the stone-free rate after transurethral ureterolithotripsy was developed. It has a reasonable predictive accuracy, and in combination with extracorporeal shock wave lithotripsy nomograms, it might be useful for deciding treatment methods.

Population pharmacokinetics of mycophenolic acid and dose optimization with limited sampling strategy in liver transplant children

The aims were to estimate the mycophenolic acid (MPA) population pharmacokinetic parameters in paediatric liver transplant recipients, to identify the factors affecting MPA pharmacokinetics and to develop a limited sampling strategy to estimate individual MPA AUC(0,12 h). Twenty-eight children, 1.1 to 18.0 years old, received oral mycophenolate mofetil (MMF) therapy combined with either tacrolimus (n= 23) or ciclosporin (n= 5). The population parameters were estimated from a model-building set of 16 intensive pharmacokinetic datasets obtained from 16 children. The data were analyzed by nonlinear mixed effect modelling, using a one compartment model with first order absorption and first order elimination and random effects on the absorption rate (k(a)), the apparent volume of distribution (V/F) and apparent clearance (CL/F). Two covariates, time since transplantation (≤ and >6 months) and age affected MPA pharmacokinetics. k(a), estimated at 1.7 h(-1) at age 8.7 years, exhibited large interindividual variability (308%). V/F, estimated at 64.7 l, increased about 2.3 times in children during the immediate post transplantation period. This increase was due to the increase in the unbound MPA fraction caused by the low albumin concentration. CL/F was estimated at 12.7 l h(-1). To estimate individual AUC(0,12 h), the pharmacokinetic parameters obtained with the final model, including covariates, were coded in Adapt II(®) software, using the Bayesian approach. The AUC(0,12 h) estimated from concentrations measured 0, 1 and 4 h after administration of MMF did not differ from reference values. This study allowed the estimation of the population pharmacokinetic MPA parameters. A simple sampling procedure is suggested to help to optimize pediatric patient care.

P4-10-05: Improved Breast Cancer Risk Assessment in Biopsied Women Using the Polyfactorial Model OncoVue®.

Abstract Background Regardless of the result, women that have undergone a breast biopsy are considered at increased risk due to the clinical circumstances leading to the need for the procedure. Both the number and outcome of biopsy are considered in the widely used Breast Cancer Risk Assessment Tool (BCRAT) also known as the Gail model. Accurately estimating individualized risk of developing breast cancer is useful for early detection and cancer prevention. Although the BCRAT has been demonstrated to accurately estimate the number of breast cancers likely to emerge in a population of women seeking regular mammography screening, it does not produce accurate individualized risk estimates for patient counseling. We have simultaneously analyzed gene polymorphism and clinical factor data in breast cancer cases and matched controls to develop a polyfactorial risk model (OncoVue) to improve estimation of individual risk. In three independent patient populations, OncoVue has been shown to significantly outperform both the BCRAT and composite risk scores produced by combining GWAS SNP risks with the BCRAT. Here we have characterized the performance of OncoVue in stratifying risk in women that have had one or more breast biopsies. Materials and Methods: Risk scores were analyzed for participants ranging in age from 35 to 89 for a subset of participants that had enrolled in a larger case-control study conducted in six distinct geographic regions of the United States. The current study focused on the analysis of participants that had reported one or more biopsies (cases prior to diagnosis/controls at time of enrollment) amounting to 1265 Caucasian women (537 cases and 728 controls) in a model building set and 303 women in an independent validation set (134 cases and 169 controls). DNAs were genotyped for 22 SNP variants and genotype information was combined with clinical risk factor information to calculate the risk scores for the individual participants. Clinical factor information was also used to calculate BCRAT risk scores. The performance of OncoVue was examined in comparison to the BCRAT alone. Results: For both models, positive likelihood ratios (PLR) were calculated as the proportion of patients with breast cancer with an elevated lifetime risk estimate (≥20%) divided by the proportion of disease-free individuals with an elevated risk estimate. In both the model building and validation sets, OncoVue exhibited approximately a 2.0-fold improvement compared to the BCRAT in more accurately assigning elevated risk estimates to breast cancer cases. In these women that are already considered at increased risk because of a history of biopsy, the observed level of improved performance of OncoVue was similar to that in our previous overall studies. Conclusions: The OncoVue polyfactorial risk model incorporating both genetics and clinical factors improves on individualized breast cancer risk estimation compared to the BCRAT which uses only clinical factors. The performance in biopsied women further supports the potential utility of OncoVue for directing prevention and screening decisions. Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr P4-10-05.

Assessing the added value of breast tumor markers in breast cancer genetic risk prediction model BRCAPRO.

164 Background: BRCAPRO is a widely used genetic risk prediction model for estimating the carrier probabilities of mutations in BRCA1/2 genes. BRCAPRO has been enhanced to utilize information on molecular markers ER, PR, CK5/6, and CK14. However, no independent validation study on the utility of these markers in risk prediction exists to support using these in actual clinical settings. Further, an important predictive and prognostic marker for breast cancer, Her-2/neu (Her2) is not utilized in BRCAPRO. Therefore, the aim of our study was to: 1) incorporate Her2 in BRCAPRO; 2) conduct a validation study of the markers. Methods: Patients with breast cancer at the UT M. D. Anderson Cancer Center’s breast clinic who were referred for genetic evaluation were included. Separate sets of cohort were used for model building with Her2 and validation to avoid bias. This study was approved by the IRB. For the model building, we estimated the joint probabilities of ER and Her2 status for carriers and non-carriers of BRCA1/2 mutations. For the validation, BRCAPRO was run at two settings: 1) no marker data used and 2) ER/PR used. We calculated the Area Under the receiving operator characteristic Curve (AUC) using the probabilities of carrying any of BRCA1 or BRCA2 and conditional probabilities of carrying BRCA1 (CondBRCA1) and BRCA2 (CondBRCA2) given a proband is carrier. Results: The model-building set for Her2 was based on 409 probands and validation set on 796 probands wherein 23% of the probands were carriers. In the model-building step, we found that joint consideration of Her2 and ER/PR is useful in discriminating between carriers and non-carriers in some subgroups, e.g., a proband with ER-, Her2+ is much more likely to be a non-carrier than a carrier. In the validation step, the AUC for CondBRCA1 and CondBRCA2 improved substantially when ER/PR was used. We are in the process of coding Her2 in BRCAPRO and then validating its utility. Conclusions: Breast tumor markers are useful for prediction of BRCA1/2 mutation status in the BRCAPRO model. ER/PR helps discriminate between BRCA1 and BRCA2 mutation carriers. In our ongoing validation study Her2 is expected to improve discrimination between carriers and non-carriers in certain sub-groups.

A Novel Clinical Instrument for Predicting Delayed Recovery After Musculoskeletal Injuries

Early identification of patients at risk for delayed recovery after an injury is important to effectively target rehabilitation. This study presents a new instrument, the Prediction of Prolonged Self-Perceived Recovery After Musculoskeletal Injuries questionnaire (PPS), for prediction of self-perceived nonrecovery after musculoskeletal injuries. On the basis of a historic cohort (model building set, n = 557), we constructed the PPS consisting of two demographic variables (educational level and working status), a crude injury classification, and patient-rated physical and mental complaints during the acute phase of the injury. We evaluated the PPS's ability to predict self-perceived nonrecovery at 6 months in a new group of patients with minor musculoskeletal traffic-related injuries (validation set, n = 279). Our findings demonstrate that the PPS foresees an unfavorable course with a greater accuracy than prediction based exclusively on information about the injury. The overall percentage of correct predictions in the model building set was 77%. The overall percentage of correct predictions in the validation set was 67%. The sensitivity and specificity in relation to nonrecovery at 6 months was 55% and 73%, respectively. This is the first prospective clinical study in which an instrument is used for prediction purpose. On the basis of our results, we think that the PPS, even if not fully developed, can be used by clinicians as a tool for early identification of patients at risk for delayed recovery after trauma. A nonnegligible proportion of the patients who would benefit from additional rehabilitation are missed by the instrument in its present form. Further research is needed to verify our results.

Using the Daily Record of Severity of Problems as a Screening Instrument for Premenstrual Syndrome

To assess symptom ratings on the first day of menses to identify women at high risk of clinically significant premenstrual syndrome (PMS) who should undergo further evaluation. A cohort of 697 women kept daily symptom ratings using the Daily Record of Severity of Problems (DRSP). The DRSP includes 21 symptom items grouped within 11 domains. DRSP scores on the first day of menses were calculated using the sum of all 21 items (standard method), the sum of the highest rated items within each domain (alternative method), and the sum of seven items derived from modeling. Seventy percent of the study sample was randomly assigned into a model-building set to identify optimal cutoff scores for PMS screening. The remaining 30% comprised a testing set used to compare PMS screening results to a PMS diagnosis based on two cycles of daily DRSP ratings. Of the initial study sample, 388 participants (55.7%) completed two cycles of daily ratings. The prevalence of PMS was 30.4%. In the model-building set, the positive and negative predictive values of the 21-item DRSP scores were 53.8% and 83.4% using the standard method and 52.7% and 84.0% using the alternative method. Corresponding values were 55.0% and 84.9% for an abbreviated seven-item DRSP version. These results were confirmed in the testing set. The DRSP administered on the first day of menses is an acceptable screening instrument to identify women who may have PMS.