Highest Akaike Information Criterion Research Articles

Lower Prognostic Performance of the International Metabolic Prognostic Index Compared to the Conventional IPI and NCCN-IPI in a Real-World Cohort of Diffuse Large B-Cell Lymphoma

Introduction: The international metabolic prognostic index (IMPI) is a recently proposed model that shows improved outcome prediction for patients with diffuse large B-cell lymphoma (DLBCL) compared to the International Prognostic Index (IPI) (Mikhaeel et al.). IMPI combines total metabolic tumor volume (TMTV) as continuous variables with clinical factors, allowing individualized prognosis estimation. However, the applicability of IMPI to routine clinical practice remains uncertain since the model construction was based on data from clinical trials. Furthermore, the comparison of the model performance between IMPI and the National Comprehensive Cancer Network IPI (NCCN-IPI) is not reported so far. Methods: We performed retrospective analysis of consecutive patients with newly diagnosed DLBCL who were initially treated with R-CHOP (rituximab plus cyclophosphamide, doxorubicin, vincristine, and prednisone) at Kameda Medical Center between 2006 and 2020. Patients who had no fluorodeoxyglucose-avid lesions were excluded. TMTV was defined as the volume of lymphoma-associated lesions with a standardized uptake value of ≥4 as the absolute threshold (Metavol, Hokkaido University, Japan). According to the original study (Mikhaeel et al.), IMPI was calculated using the following formula: 0.003077330 × (MTV lower than the median of 307.9ml) - 0.002761985 × (MTV higher than the median of 307.9ml) + 0.008092449 × Age -0.114645415 × Stage 2 + 0.281141117 × Stage 3 + 0.322247142 × Stage 4. The receiver operating curve (ROC) for predicting 5-year survival was compared among IMPI-, IPI-, and NCCN-IPI-based classifications. The predictive performance of these three systems was evaluated using the Akaike information criterion (AIC) and Harrell's concordance index (c-index). Results: A total of 314 patients were included in the study. The median age was 72 years old (interquartile range [IQR] 64-79), which was younger than the original data (median 62, IQR 51-70). Regarding the distributions of IPI risk groups, this cohort had a lower number of low-risk patients (27.0% vs. 32.4%) and a higher number of high-risk patients (23.9% vs. 19.5%). The median TMTV was 168 (IQR 47-530), which was approximately two times lower than that in the original study (median 307, IQR 77-838). After calculating and ranking the individual IMPI scores, nearly half of the patients (n=147, 46.8%) were reclassified into different risk groups from the original IPI categories (Figure 1); 76 (51.7%) were upstaged, and 71 (48.3%) were downstaged. Of note, the reclassification resulted in a significantly better OS prediction only for the IPI high-risk group (5-year OS rate, IMPI high-risk vs. IMPI not high-risk, 41.9% vs. 59%, P =0.042), while it did not for the IPI low-, low-intermediate-, and high-intermediate-risk groups. The area under the curve (AUC) from the ROC curve predicting 5-year survival was not significantly different between IMPI- (AUC 0.65) and either IPI- (AUC 0.67, P =0.42) or NCCN-IPI-categories (AUC 0.68, P=0.25). Finally, the predictive performance and model fitness were compared among these three indices. IMPI had the highest AIC (1117.68) and the lowest c-index (0.643) for OS prediction, whereas NCCNIPI yielded the best performance (AIC 1097.66 and c-index 0.677) (Table 1). Conclusions: We demonstrated that the IMPI model did not outperform the conventional IPI and NCCN-IPI systems in a real-world cohort of DLBCL patients. The difference in age and tumor burden, which are the major components of IMPI, may lead to poorer prognostic performance. Further large-scale studies are warranted to validate the prognostic value of IMPI outside of clinical trials.

Comparison of survival analysis approaches to modelling age at first sex among youth in Kisesa Tanzania.

Many studies analyze sexual and reproductive event data using descriptive life tables. Survival analysis has better power to estimate factors associated with age at first sex (AFS), but proportional hazards models may not be right model to use. This study used accelerated failure time (AFT) models, restricted Mean Survival time model (RMST) models, with semi and non-parametric methods to assess age at first sex (AFS), factors associated with AFS, and verify underlying assumptions for each analysis. Self-reported sexual debut data was used from respondents 15-24 years in eight cross-sectional surveys between 1994-2016, and from adolescents' survey in an observational community study (2019-2020) in northwest Tanzania. Median AFS was estimated in each survey using non-parametric and parametric models. Cox regression, AFT parametric models (exponential, gamma, generalized gamma, Gompertz, Weibull, log-normal and log-logistic), and RMST were used to estimate and identify factors associated with AFS. The models were compared using Akaike information criterion (AIC) and Bayesian information criterion (BIC), where lower values represent a better model fit. The results showed that in every survey, the Cox regression model had higher AIC and BIC compared to the other models. Overall, AFT had the best fit in every survey round. The estimated median AFS using the parametric and non-parametric methods were close. In the adolescent survey, log-logistic AFT showed that females and those attending secondary and higher education level had a longer time to first sex (Time ratio (TR) = 1.03; 95% CI: 1.01-1.06, TR = 1.05; 95% CI: 1.02-1.08, respectively) compared to males and those who reported not being in school. Cell phone ownership (TR = 0.94, 95% CI: 0.91-0.96), alcohol consumption (TR = 0.88; 95% CI: 0.84-0.93), and employed adolescents (TR = 0.95, 95% CI: 0.92-0.98) shortened time to first sex. The AFT model is better than Cox PH model in estimating AFS among the young population.

The feasibility of a Bayesian network model to assess the probability of simultaneous symptoms in patients with advanced cancer

Although patients with advanced cancer often experience multiple symptoms simultaneously, clinicians usually focus on symptoms that are volunteered by patients during regular history-taking. We aimed to evaluate the feasibility of a Bayesian network (BN) model to predict the presence of simultaneous symptoms, based on the presence of other symptoms. Our goal is to help clinicians prioritize which symptoms to assess. Patient-reported severity of 11 symptoms (scale 0–10) was measured using an adapted Edmonton Symptom Assessment Scale (ESAS) in a national cross-sectional survey among advanced cancer patients. Scores were dichotomized (< 4 and ≥ 4). Using fourfold cross validation, the prediction error of 9 BN algorithms was estimated (Akaike information criterion (AIC). The model with the highest AIC was evaluated. Model predictive performance was assessed per symptom; an area under curve (AUC) of ≥ 0.65 was considered satisfactory. Model calibration compared predicted and observed probabilities; > 10% difference was considered inaccurate. Symptom scores of 532 patients were collected. A symptom score ≥ 4 was most prevalent for fatigue (64.7%). AUCs varied between 0.60 and 0.78, with satisfactory AUCs for 8/11 symptoms. Calibration was accurate for 101/110 predicted conditional probabilities. Whether a patient experienced fatigue was directly associated with experiencing 7 other symptoms. For example, in the absence or presence of fatigue, the model predicted a 8.6% and 33.1% probability of experiencing anxiety, respectively. It is feasible to use BN development for prioritizing symptom assessment. Fatigue seems most eligble to serve as a starting symptom for predicting the probability of experiencing simultaneous symptoms.

Modified staging system for gastric neuroendocrine carcinoma based on American Joint Committee on Cancer and European Neuroendocrine Tumor Society systems.

The prognostic values of the AJCC staging system for gastric cancer (GC-AJCC), the AJCC staging system for gastric neuroendocrine tumours (NET-AJCC) and the European Neuroendocrine Tumor Society (ENETS) system for gastric neuroendocrine carcinoma and mixed adenoneuroendocrine carcinoma (MA)NEC remain controversial. Data on patients with (MA)NEC from 21 centres in China were analysed. Different staging systems were evaluated by performing Kaplan-Meier survival analysis and calculating the concordance index (C-index) and Akaike information criterion (AIC). Based on three existing systems, a modified staging system (mTNM) was developed. A total of 871 patients were included. In the GC-AJCC system, an overlap was noticed for pT2 and pT3 categories. Patients with stage IIIC disease had a similar prognosis to those with stage IV disease. The pT categories of the NET-AJCC system had a lower C-index and higher AIC than those of the other systems. In the ENETS system, there was a low proportion (0·2 per cent) of patients with stage IIIA and a high proportion (67·6 per cent) of stage IIIB disease. The mTNM system adopted the NET-AJCC pT and GC-AJCC pN and pM definitions, and was developed based on the ENETS stage definitions. The proportion of patients in each stage was better distributed and the mTNM system showed improved prognostic performance in predicting overall and disease-free survival. The mTNM system offers more accurate prognostic value for gastric (MA)NEC than the AJCC or ENETS staging systems.

Validation of the Prognostic Significance of the Prognostic Stage Group According to the Eighth Edition of American Cancer Joint Committee on Cancer Staging System in Triple-Negative Breast Cancer: An Analysis From Surveillance, Epidemiology, and End Results 18 Database

BackgroundThe eighth edition of the American Cancer Joint Committee on Cancer (AJCC) staging system for breast cancer put forward the prognostic stage groups for the first time based on the traditional anatomic tumor-node-metastasis staging system. Our study intends to validate the predictive significance of the eighth edition staging system in triple-negative breast cancer (TNBC) patients. Materials and methodsWe collected and accessed 26,589 eligible cases of TNBC from the Surveillance, Epidemiology, and End Results database (2010-2015) and reclassified the patient cohort according to the eighth edition of the AJCC staging system into anatomic and prognostic stages. ResultsThe results showed that more than half of the patients upstaged in the prognostic stage when compared with the anatomic stage. By comparing with the anatomic stage, the prognostic stage had a higher likelihood ratio and linear trend χ2 values. The prognostic stage group also had higher Akaike information criterion and Bayesian information criterion values than the anatomic stage group. ConclusionsThe prognostic staging system in TNBC patients performs more optimistic prognostic stratification and predictability than the anatomic staging system. Moreover, the latest AJCC staging system has a milestone importance to the history of breast cancer staging system.

A Novel Prognostic Model Including Cytogenetic and Molecular Data in Patients with Primary and Post-Essential Thrombocythemia (ET)/Polycythemia Vera (PV) Myelofibrosis (MF)

Introduction: MF is a Philadelphia-negative myeloproliferative neoplasm (Ph-negative MPN) with an heterogeneous outcome. In 2009, Cervantes et al. published the International Prognostic Score System (IPSS) to better determine outcomes in this disease. In the last decade, several recurrently mutated genes have been described in MF, some of them associated with prognostic impact in survival. We propose a novel prognostic score that incorporates molecular and cytogenetic data in patients with MF.Methods: We analyzed clinical, cytogenetic and molecular data from 623 patients with a diagnosis of primary MF (N=445), post-PV MF (N=109) and post-ET MF (N=69). Data was extracted from medical records at time of sample collection for analysis. Mutation data was obtained by next-generation sequencing analysis, performed with either paired tumor-normal whole exome sequencing (N=46) or selected gene panel for genes associated with myeloid malignancies (N=577). The following 16 genes were analyzed in all 623 patients and were considered as the common denominator for analysis: ASXL1, CALR, DNMT3A, EZH2, FLT3, IDH1, IDH2, JAK2, KIT, KRAS, MPL, NRAS, RUNX1, TET2, TP53, WT1. RAS mutations were considered as oncogenic mutations in NRAS and/or KRAS. Molecular high risk (MHR) mutations were considered as mutations in any one of the 4 genes: ASXL1, EZH2, IDH1, IDH2 (SRSF2 mutations were not included since they were not evaluated in all cases). Cytogenetic data was stratified into 4 risk categories (based on Tam et al, Blood 2009): (1) Diploid; (2) Del(13q)/Del(20q)/Trisomy 9; (3) Abnormalities of chromosomes 5, 7, 17 and complex karyotype; (4) Other abnormalities. To develop the model, the data was split into a training dataset (N=434) and a test dataset (N=189). Variables initially included in the initial training model were those with a p-value<0.05 in the univariate analysis; subsequently, this initial model was internally validated with a bootstrap procedure for stepwise backwards selection and determination of variables to be included the final model. Score points were assigned based on the value of the hazard ratio (HR). The final model was externally validated in the test dataset. Akaike information criterion (AIC) was calculated on the final model and on the Dynamic IPSS (DIPSS) model on the full dataset, and the best model was considered the one with the lowest AIC value and the highest AIC weight.Results: In the training cohort, after a median follow-up of 30.8 months, there were 176 deaths (40.5%). The initial variables included in the multivariate Cox model were: age (>65 years), hemoglobin (<10 g/dL), white blood cell count (WBC; >25x109/L), peripheral blood blasts (>1%), presence of constitutional symptoms, sex (male vs female), platelet count (<100x109/L), type of JAK-STAT driver mutation (CALR vs JAK2 vs MPL vs Triple-negative), karyotype risk category, presence of MHR mutations and presence of RAS mutations. After bootstrapping, variables sex, peripheral blood blasts and WBC were excluded, and the final model with associated HRs is presented on table 1. Since most HRs were near 1, 1-point was given to each variable, with the exception of Triple-negative molecular status and RAS mutations (2 points each). We grouped patients based on similarity of HR into three risk categories: low (0-2 points; N=243), intermediate (3-4 points, N=141) and high (5+ points, N=50). These 3 groups were associated with significantly different survival outcomes (Figure 1): 3-years OS was 71% (95% CI 64-78%), 38% (95% CI 29-50%) and 18% (95% CI 9-37%) for low, intermediate and high risk groups (P=1.33e-13). Validating the model on the test dataset revealed significantly differences in OS among the 3 risk categories (Figure 2): 74% (95% CI 63-86%), 50% (95% CI 36-68%) and 16% (95% CI 6-39%) (P=1.42e-11). Analyzing the full dataset, the novel score model had a lower AIC than the DIPSS model (2713 vs 2722) and a higher AIC weight (0.99 vs. 0.01).Conclusion: Incorporating molecular data, including MHR mutations, karyotype and RAS mutations leads to the development of an improved prognostic model. Further validation of this model in other cohorts is necessary. The impact of novel therapies should be evaluated among the different risk categories. [Display omitted] DisclosuresRampal:Stemline: Research Funding; Celgene: Honoraria; Constellation: Research Funding; Incyte: Honoraria, Research Funding; Jazz: Consultancy, Honoraria. Verstovsek:Celgene: Membership on an entity's Board of Directors or advisory committees; Italfarmaco: Membership on an entity's Board of Directors or advisory committees; Novartis: Membership on an entity's Board of Directors or advisory committees, Research Funding, Speakers Bureau; Incyte: Consultancy.

Common mechanisms for guidance efficiency of descending Atlantic salmon smolts in small and large hydroelectric power plants

AbstractDams and turbines associated with hydroelectric power plants (HEP) disrupt connectivity by affecting fish movement and survival. There has been an increasing focus on measures to facilitate downstream migration at HEPs. The fish guidance efficacy (FGE) of downstream mitigation measures largely remains suboptimal and calls for development of knowledge on factors influencing FGE. In this study, we analyse 6 years of wild Atlantic salmon smolt passive integrated transponder (PIT)‐telemetry data (N = 1,498) from a neighbouring small‐ and a large‐scale HEP. Timing of the smolt migration period was significantly different between the two rivers. Thus, river‐specific smolt‐run timing is imperative for proper measures management in regulated rivers aiming at maximizing smolt‐descent survival. A generalized linear model including additive effects of relative bypass discharge and scaled river discharge on the FGE for descending smolts received highest Akaike's information criterion support in the data and explained 74.2% of the FGE variation. This model, including no river effect, predicted high FGE (up to 90%) at low river flow (≤30% of HEP maximum capacity) when 7% of the water is allocated through the bypass. Many run‐of‐the‐river HEPs have highly variable river flow during the smolt‐run period. Our model suggests that these HEPs could utilize their manoeuvre flexibility to obtain water allocation routines between bypass and turbines that optimize both FGE and hydroelectric production.

Validation of the 8th Edition of the UICC/AJCC Staging System for Nasopharyngeal Carcinoma From Endemic Areas in the Intensity-Modulated Radiotherapy Era.

Background: In this study, we evaluated the 8th edition of the Union for International Cancer Control (UICC)/AJCC staging system for nasopharyngeal carcinoma (NPC) in an endemic area, with the aim of validating its applicability and providing further information for future refinements. Methods: A total of 1,790 patients with newly diagnosed, non-distant metastatic, histologically proven NPC treated with intensity-modulated radiotherapy (IMRT) were retrospectively reviewed. The performance of various staging systems was compared using the Akaike information criterion (AIC) and Harrell's concordance index (c-index). Results: For N (node) category, the survival curves of different groups according to the 8th edition were well-separated, and the prognostic model predicted outcomes fairly well. The 8th edition had higher AIC and c-index values for all end points than the 7th edition. However, probably due to the improved locoregional control provided by IMRT, the survival curves for T2 and T3 almost overlapped, without significant differences in locoregional failure-free survival (P=.606) and disease-free survival (P=.735). Due to the difficultly of differentiating T2 and T3, the AIC and c-index values were similar for the T categories of the 7th and 8th editions. Similarly, the overall survival and disease-free survival curves for stage II and III disease were not clearly separated for either the 8th or 7th editions. Conclusions: The 8th edition of the UICC/AJCC staging system for NPC enables more accurate prediction of treatment outcomes. However, several limitations need to be addressed in future editions, and it would be reasonable to further optimize the T category classification.

Comparison of Risk Adjustment Methods in Patients with Liver Disease Using Electronic Medical Record

ABSTRACTObjectiveRisk adjustment methods are widely used to compare quality of care or predict health outcome, but the optimal approach is unclear for liver disease. This study is to compare the performance of common risk adjustment methods for predicting in-hospital mortality in patients with liver disease using Electronic Medical Record (EMR).&#x0D; ApproachThe EMR data was derived from Beijing YouAn Hospital between 2010 and 2015. 85,526 EMRs were included. Previously developed and validated automated EMR case definitions were applied to define the conditions including primary liver cancer, cirrhosis and other conditions included in Charlson, Elixhauser comorbidity algorithms, Child-Turcotte-Pugh (CTP) score and Model for End-Stage Liver Disease (MELD). Logistic regression was conducted and C-statistic was obtained to compare the performance of the different methods for predicting in-hospital mortality. To eliminate the effect of the model complexity on model performance, we compared Akaike Information Criterion (AIC) of different methods (smaller AIC is better).&#x0D; ResultIn total, we included three liver diseases cohort: 7,178 Primary Liver Cancer (PLC) patients, 11,121 cirrhosis patients and 7,298 cirrhosis without PLC patient. For PLC cohort, C-statistics of these compared indexes ranged from 0.72 to 0.84; AIC was between 4312.3 and 5048.4. For cirrhosis cohort, C-statistics of these compared indexes ranged from 0.73 to 0.83; AIC was between 4952.1 and 5788.2. For cirrhosis without PLC cohort, C-statistics ranged from 0.73 to 0.84; AIC was between 2608.3 and 3240.5. It was consistent across the three cohorts that MELD + sodium (MELD_Na) score (a variant of MELD score) had the highest C-statistic and lowest AIC; CTP had the lowest C-statistic and highest AIC. Integrating Charlson Comorbidity to MELD_Na, C-statistic improved to 0.86 and AIC reduced.&#x0D; ConclusionAmong the compared risk adjustment methods, MELD_Na performed best for predicting in-hospital mortality among patients with PLC or cirrhosis using Chinese EMRs. Adding clinical information to comorbidity algorithms improved the performance of the model.

Comparison of Statistical Models for the Estimation of Age at Death Using Subjective Adult Human Dental Indicators

Objective: The main objective of this paper is the estimation of age at death using subjective dental data. This is particularly useful in developing and under developed countries. Methods: This study provides a framework for the estimation of age at death using very subjective measurements of the teeth using (i) Generalized Linear Models (GLMs) and (ii) Generalized Additive Models (GAMs). These predictors of age were all ordinal in nature. A dataset comprising measurements taken on 71 maxillary incisors from different individuals at the time of their death was used. A comparison of two models – the Gamma GLM and the Gamma GAM is used to illustrate the flexibility of this method and the predictive power of the statistical modelling process. Results: The study showed the effectiveness of the models through the Akaike Information Criterion (AIC) as well as the proportion of correct predictions within each of the age groups. The Gamma GAM actually had the higher AIC but the better predictive values within the age groups. Conclusion: Statistical modelling caters for the types of data and can give reasonable predictions of age at death.

Stepwise Logistic Regression

In Response: We thank Dr. Arunajadai for his comments about the statistical simulations in our editorial (text NLP, algorithm WMB) demonstrating the perils of stepwise logistic regression.1 This allows us to clarify an ambiguity in the nomenclature of the stepwise automatic variable selection algorithm. Correctly specified, the algorithm should be described as either stepwise forward selection, stepwise backward elimination, or stepwise with forward selection and/or backward elimination; however, the word stepwise itself is also commonly used to refer to any of the three variants or to just the third variant. Arunajadai2 has correctly stated that our particular simulations used the stepwise backward elimination variant. Our simulations used randomly created covariates to demonstrate how commonly there was the creation of spurious associations by stepwise modeling (backward elimination variant). Dr. Arunajadai has also provided R software code to perform the other two variants; he reports that there were no spurious associations with no covariate significant at P < 0.05 using either the forward selection or the forward selection/ backward elimination variants. In his code, Arunajadai estimates a mean intercept model object, i.e., “fit <- glm(y ∼ 1, data = w, family = binomial),” for submission to the stepwise function. The submission of a mean intercept model to the stepwise process cannot identify any association, true or spurious. When a full (all covariates) model, i.e., “fit <- glm(y ∼., data = w, family = binomial)” is used, all three variants have qualitatively the same results of numerous spurious associations (appendix available at www.anesthesia-analgesia.org). The inclusion of noise variables during stepwise modeling regardless of the variant has been demonstrated elsewhere.3–5 Dr. Arunajadai also raised the very interesting question of which information criterion should be used at each step for adding or removing a covariate; he advocates the Bayesian Information Criterion (BIC) in contrast to the Akaike Information Criterion (AIC) used in our simulation. Both the AIC and the BIC are indexes in which twice the negative maximized log likelihood of the model fit is penalized by subtracting either twice the number of model parameters (AIC) or the number of model parameters multiplied by the log of the sample size (BIC). Of the candidate models possible, the model with the higher AIC or higher BIC is favored. As Arunajadai noted, the BIC is more heavily penalized and will produce more parsimonious models (fewer significant covariates). However, there is a competition in choosing between AIC and BIC; the AIC will yield optimal regression estimation while the BIC represents consistent model identification rules. It is not possible to create models with the properties favored by both the AIC and the BIC.6 Using the BIC index in our simulation still produces spurious associations. Automatic variable selection via a stepwise process is a hazardous undertaking. As J. B. Copas3 humorously noted, “If you torture the data for long enough, in the end they will confess …. What more brutal torture can there be than subset selection? The data will always confess, and the confession will usually be wrong.” Nathan L. Pace, MD, MStat Department of Anesthesiology University of Utah Salt Lake City, Utah William M. Briggs, PhD Department of Emergency Medicine New York Methodist Hospital Brooklyn, New York [email protected]

Development of a Sightability Model for Low‐Density Elk Populations in Ontario, Canada

ABSTRACT The status of recolonizing elk (Cervus elaphus) populations in Ontario, Canada, is unclear and there is a need for effective population survey methods that can be applied locally. We sought to develop a sightability model that could account for both low densities of elk and dense forest cover in elk‐release areas in Ontario. We corrected winter aerial survey counts for sightability based on radiocollared animals known to be within observable distance of the aircraft. The multivariate model with the highest Akaike's Information Criterion corrected for sample size weight (wi = 0.427) revealed that elk group size, elk activity, dominant tree type, percent canopy cover, and percent conifer cover were significant predictors of elk sightability. The group‐size effect indicated that odds of sighting an elk increased by 1.353 (95% CI = 0.874‐3.689) for every additional elk. Standing elk were 5.033 (95% CI = 0.936‐15.541) times more likely to be observed than were resting elk, and those located in conifer cover were 0.013 (95% CI = 0.001‐0.278) times less likely to be sighted than elk in deciduous cover. Furthermore, elk located in &gt;50% canopy cover and &gt;50% conifer cover were 0.041 (95% CI = 0.003‐0.619) times and 0.484 (95% CI = 0.024‐9.721) times less likely to be sighted than elk in more open habitat, respectively. During model validation, observers detected 79% (113/143) of known elk in any given area, and population and sightability model predictions (±90% CI) overlapped with the population estimate, implying that our predictive model was robust. Unsurprisingly, large groups of elk in open habitat increased model precision, which highlights difficulties of counting Ontario elk in their northern range. We conclude that our model provided increased reliability for estimating elk numbers in Ontario compared to existing methods, and that the estimator may be useful in other areas where elk density is low and sightability is poor due to dense forest cover.

External Prognostic Validations and Comparisons of Age- and Gender-Adjusted Exercise Capacity Predictions

The purpose of this study was to externally validate the prognostic value of age- and gender-based nomograms and categorical definitions of impaired exercise capacity (EC). Exercise capacity predicts death, but its use in routine clinical practice is hampered by its close correlation with age and gender. For a median of 5 years, we followed 22,275 patients without known heart disease who underwent symptom-limited stress testing. Models for predicted or impaired EC were identified by literature search. Gender-specific multivariable proportional hazards models were constructed. Four methods were used to assess validity: Akaike Information Criterion (AIC), right-censored c-index in 100 out-of-bootstrap samples, the Nagelkerke Index R2, and calculation of calibration error in 100 bootstrap samples. There were 646 and 430 deaths in 13,098 men and 9,177 women, respectively. Of the 7 models tested in men, a model based on a Veterans Affairs cohort (predicted metabolic equivalents [METs] = 18 - [0.15 x age]) had the highest AIC and R2. In women, a model based on the St. James Take Heart Project (predicted METs = 14.7 - [0.13 x age]) performed best. Categorical definitions of fitness performed less well. Even after accounting for age and gender, there was still an important interaction with age, whereby predicted EC was a weaker predictor in older subjects (p for interaction <0.001 in men and 0.003 in women). Several methods describe EC accounting for age and gender-related differences, but their ability to predict mortality differ. Simple cutoff values fail to fully describe EC's strong predictive value.