SEARCH · PubMed Health

Results for “predictive modelling”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,711 records · Page 95Linked to original sources

PubMed2005

A bedside prediction-scoring model for late-onset neonatal sepsis.

OBJECTIVE: Insufficient tools for bedside prediction of late-onset neonatal sepsis (LNS) initiated this study. The objective was to develop and validate a simple prediction-scoring model for LNS defined as culture-proven sepsis occurring 72 hours after birth. METHODS: The study was performed at a university hospital in Bangkok. The derivation phase included medical records of 1870 neonates, randomly selected from 9347 records of neonates who had been hospitalized for >72 hours during 1998 to 2000, of which 1824 records were available. In all, 100 neonates were clinically suspected of sepsis and 17 had proven LNS. The validation phase included 73 neonates suspected of having sepsis during July 2002 to June 2003 and 25 who had LNS. Weighted coefficients from Cox's proportional hazards model and receiver-operating-characteristic (ROC) curve analysis were used. RESULTS: The incidence density of LNS was 17/11355 (1.5/1000) person-days. A scoring model was developed and consisted of the following: hypotension (score 4), abnormal body temperature (score 3), respiratory insufficiency (score 2), neutrophil band form fraction >1% (score 2), platelet count <150 x 10(3)/microl (score 2), and umbilical venous catheterization (1 to 7 or >7 days; score 2 or 4). The area under the ROC curves for prediction of LNS in a neonate suspected of sepsis in each of the two phases was 0.85 and 0.80, respectively (p=0.436). The mean probabilities of LNS were approximately 0.10 (low risk) for scores from 0 to 3; 0.50 (intermediate risk) for scores from 4 to 6; and 0.70 (high risk) for scores > or =7. CONCLUSION: A simple prediction-scoring model for LNS was developed. Validation of the scores suggested good diagnostic performance.

Cohort Studies↗

PubMed2002

A design of two simple models to predict PCDD/F concentrations in vegetation and soils.

The vegetation and soil levels of the 17 polychlorinated dibenzo-p-dioxin and dibenzofuran (PCDD/F) toxic congeners were calculated by means of a vegetation and a soil model, respectively. Both models predicted the levels of the 17 PCDD/F congeners in quite good agreement with the observed results although the soil model was more accurate than the vegetation model. Four different pathways of contribution to the vegetation concentrations were taken into account: vapour-phase absorption, dry particle deposition, wet particle deposition and uptake by root. The most important pathway was the vapour-phase absorption and the less was the uptake by root. In the soils model four pathways were considered: background soil concentration, dry particle deposition, wet particle deposition and uptake by root. After the background concentration, the most important pathway was the wet deposition.

Absorption↗

PubMed2000

Chronic addiction relapse treatment: a study of the effectiveness of the high-risk identification and prediction treatment model. Part I.

The high-risk identification and prediction treatment model (HRIPTM) was developed to investigate physiological and psychological conditions associated with the concept of negative emotional state, and to examine the origin and evolution of this condition in relation to addiction relapse. An in-depth case study sample of n=30 was conducted within the context of an intensive out-patient addiction treatment program based on the HRIPTM. The HRIPTM investigates the nature of problematic repetitive behavior patterns originating in the childhood developmental period, the individual nature of negative emotional state in relation to these patterns and their evolution, and their influence on addiction vulnerability and addiction relapse. The preliminary study, upon which this model was based, emphasized the importance of developing methods for accumulating information that pinpoints lifespan experiences associated with the origin of problematic behaviors, the level of emotional intensity factors associated with behavior pattern construction and development, and level of associated risk potential as it manifests in each individual case. Recent neuroscientific studies into the early developmental process are producing increased evidence of the enormous influence of this period on later developmental processes. Addiction, in itself a developmental process, may be more influenced by pre-addiction factors than previously assumed. This model addresses those influences. The study was conducted to determine whether or not information that arises from the HRIPTM increases the patient's and clinician's ability to address addiction relapse more effectively. As a result of completion of the HRIPTM program, the sample n=30 reported gained ability to address recovery processes more effectively. The 12-question HRIPTM Questionnaire instrument, when tabulated, revealed that of the 360 total responses, less than 2% disagreed that ability had been gained in all suggested areas defined in the questionnaire. Over 98% agreed or strongly agreed that ability had been gained in the defined areas.

Humans↗

PubMed2004

Intersection of microwire electrodes with proximal CA1 stratum-pyramidale neurons at insertion for multiunit recordings predicted by a 3-D computer model.

It is of broad interest in the context of neuronal multiunit extracellular recordings to understand electrode-tissue interactions in order to maximize the number of recordable units and to minimize experimental artifacts due to mechanical tissue alteration. Toward this goal, a computer model of microwire electrode insertion in hippocampus CA1 area was developed, firstly to provide estimates of the number of electrode-neuron intersections affecting recordable (local) neurons and, secondly, to determine optimal insertion/electrode parameters that minimize the number of intersections. The model predicts that in hippocampus CA1 area, using an electrode 50 microm in diameter, only 10% of the recordable neurons (those within 50 microm of the electrode), would remain collision free. Moreover, the model also predicts that inhibitory neurons are less prone to be intersected by the electrode, resulting in a 2 to threefold higher percentage of collision-free interneurons than expected from the relative densities of pyramidal cells and interneurons. Furthermore, the model confirms, in agreement with experimental observations, that electrode tilting with respect to the main neuronal axis increases the number of intact neurons (fourfold for a 50-microm electrode at 45 degrees when compared to 0 degrees , i.e., an insertion normal to the cell body layer).

Animals↗

PubMed2026

Development and external validation of an explainable machine learning model for predicting chronic kidney disease progression in the Korean population.

BACKGROUND: Current risk stratification models, such as the Kidney Failure Risk Equation (KFRE), exhibit variable performance across ethnic groups and fail to capture dynamic clinical trajectories. This study aimed to develop and validate a Korean-specific machine learning (ML) model for predicting chronic kidney disease (CKD) progression using an ensemble approach. METHODS: We used electronic health records from Seoul National University Hospital for model development (n = 28,209) and the Korean Genome and Epidemiology Study (KoGES) CKD cohort for external validation (n = 3,960). The primary outcome was a composite of ≥40% decline in estimated glomerular filtration rate (eGFR) or progression to end-stage renal disease within 2 years. A soft-voting ensemble of four ML algorithms (XGBoost, LightGBM, CatBoost, and Random Forest) was developed. RESULTS: The ensemble model demonstrated robust discrimination in internal validation (area under the receiver operating characteristic curve [AUROC], 0.939; 95% confidence interval [CI], 0.934-0.944), significantly exceeding the KFRE (AUROC, 0.879-0.884). External validation in the KoGES cohort showed comparable discrimination (AUROC, 0.859; 95% CI, 0.798-0.914) versus KFRE (four-variable AUROC, 0.882; 95% CI, 0.818-0.935). Shapley Additive exPlanations (SHAP) analysis identified baseline eGFR, serum creatinine, eGFR slope, albumin, and hemoglobin as key prognostic features, supporting a complementary framework using KFRE for community screening and the ML model for hospital-based risk stratification. CONCLUSION: The ensemble ML model accurately predicts short-term CKD progression in Korean patients. By incorporating longitudinal features and ensemble learning, it provides a precise alternative to Western-derived equations, particularly in tertiary care settings.

Chronic kidney failure↗

PubMed2003

Application of pharmacokinetic-pharmacodynamic modeling to predict the kinetic and dynamic effects of anti-methotrexate antibodies in mice.

We have shown that intravenous (i.v.) administration of anti-methotrexate (MTX) antibodies (AMAb) reduces the systemic exposure of intraperitoneal (i.p.) MTX therapy, and we have proposed that AMAb effects on MTX systemic exposure would allow a reduction in MTX-induced systemic toxicity (i.e., producing a desirable antagonistic effect). However, many literature reports have shown that anti-toxin antibodies occasionally demonstrate unexpected agonist-like activity, increasing the extent of toxicity induced by their ligand. In this report, we have utilized a pharmacokinetic-pharmacodynamic (PKPD) model to predict the potential of AMAb to increase or decrease the magnitude of MTX-induced body weight loss in mice. Simulations predicted that both anti-MTX immunoglobulin G (AMI) and anti-MTX Fab fragments (AMF) would lead to increases or decreases in MTX toxicity, with effects dependent on the dosing protocol used. Based on the computer simulations, two protocols were selected for in vivo evaluation of predicted agonistic or antagonistic effects. Murine monoclonal AMI and AMF were produced, purified, and characterized. Agonistic effects were tested after 24-h infusion of i.p. MTX (10 mg/kg) and i.v. administration of an equimolar dose of AMI. Antagonistic effects were tested after 72-h infusion of i.p. MTX (5 mg/kg) and i.v. infusion of an equimolar dose of AMF. Consistent with model predictions of agonist-like activity, the 24-h AMI protocol led to significantly increased animal mortality (all animals died, p < 0.005) and mean nadir weight loss (p < 0.005). Also consistent with the predictions of the PKPD model, the 72-h AMF protocol significantly decreased animal mortality and mean nadir body weight loss (p < 0.01). Thus, these studies demonstrate that agonistic and antagonistic effects of anti-toxin antibodies may be predicted through the use of an integrated PKPD model.

Animals↗

PubMedDate not supplied

Validation of models for predicting the use of health technologies.

Validation must be carried out before a model can be used confidently as a tool of managerial decision making in health care. The authors describe a bootstrap approach to validating models for predicting the utilization of four technologies used in neonatal care: measurement of blood gases (gasometry), the oxygen hood, continuous positive airway pressure (CPAP), and mechanical ventilation. These models were fitted by stepwise multiple linear regression from 20 prognostic covariates of 193 neonates. One hundred bootstrap samples were generated to validate the choices of covariates in the models based on their frequencies of selection. This approach validated the models for the oxygen hood and CPAP. The regression coefficients and standard deviations for the CPAP and oxygen hood models were estimated using 200 additional bootstrap samples. A close agreement between stepwise and bootstrap estimates was observed for both models. These results suggest that bootstrap can be useful for validating models for predicting the utilization of health technologies.

Blood Gas Analysis↗

PubMed2003

Analytical model for predicting mechanotransduction effects in engineered cardiac tissue.

Mechanochemical and mechanoelectrical signaling is imperative for cardiac organogenesis and underlies pathophysiological events. New techniques for engineering cardiac tissue allow unprecedented means of modeling these phenomena in vitro. However, experimental design is often hampered by a lack of models that can be adapted to the ideal conditions these methods allow. To address these deficiencies, we developed a mathematical model to calculate the distribution of stress and strain in fibrous cardiac tissue. The fluid-fiber-collagen model characterizes the mechanical behavior of cardiac tissue and is solved analytically for the distributions of stress and strain along the myocardial fibers. An example application of the model is presented: modeling the distribution of strains in the vicinity of an ischemic region. The ischemic region is stretched during systole, as has been shown in previous one-dimensional models. Our model predicts a complex distribution of stretch in the border zone surrounding the ischemic region and in nonischemic regions surrounding the border zone. These strain patterns may predict patterns of mechanochemical coupling that results in localized fibrosis, altered gene expression, or the mechanoelectrical signaling events that potentiate cardiac arrhythmias.

Arrhythmias, Cardiac↗

PubMed2002

Using a dynamic hydrology model to predict mosquito abundances in flood and swamp water.

We modeled surface wetness at high resolution, using a dynamic hydrology model, to predict flood and swamp water mosquito abundances. Historical meteorologic data, as well as topographic, soil, and vegetation data, were used to model surface wetness and identify potential fresh and swamp water breeding habitats in two northern New Jersey watersheds. Surface wetness was positively associated with the subsequent abundance of the dominant floodwater mosquito species, Aedes vexans, and the swamp water species, Anopheles walkeri. The subsequent abundance of Culex pipiens, a species that breeds in polluted, eutrophic waters, was negatively correlated with local modeled surface wetness. These associations permit real-time monitoring and forecasting of these floodwater and nonfloodwater species at high spatial and temporal resolution. These predictions will enable public health agencies to institute control measures before the mosquitoes emerge as adults, when their role as transmitters of disease comes into play.

Aedes↗

PubMed2004

Prediction of liver fibrosis and cirrhosis in chronic hepatitis B infection by serum proteomic fingerprinting: a pilot study.

BACKGROUND: Most noninvasive predictive models of liver fibrosis are complicated and have suboptimal sensitivity. This study was designed to identify serum proteomic signatures associated with liver fibrosis and to develop a proteome-based fingerprinting model for prediction of liver fibrosis. METHODS: Serum proteins from 46 patients with chronic hepatitis B (CHB) were profiled quantitatively on surface-enhanced laser desorption/ionization (SELDI) ProteinChip arrays. The identified liver fibrosis-associated proteomic fingerprint was used to construct an artificial neural network (ANN) model that produced a fibrosis index with a range of 0-6. The clinical value of this index was evaluated by leave-one-out cross-validation. RESULTS: Thirty SELDI proteomic features were significantly associated with the degree of fibrosis. Cross-validation showed that the ANN fibrosis indices derived from the proteomic fingerprint strongly correlated with Ishak scores (r = 0.831) and were significantly different among stages of fibrosis. ROC curve areas in predicting significant fibrosis (Ishak score >or=3) and cirrhosis (Ishak score >or=5) were 0.906 and 0.921, respectively. At 89% specificity, the sensitivity of the ANN fibrosis index in predicting fibrosis was 89%. The sensitivity for prediction increased with degree of fibrosis, achieving 100% for patients with Ishak scores >4. The accuracy for prediction of cirrhosis was also 89%. Inclusion of International Normalized Ratio, total protein, bilirubin, alanine transaminase, and hemoglobin in the ANN model improved the predictive power, giving accuracies >90% for the prediction of fibrosis and cirrhosis. CONCLUSIONS: A unique serum proteomic fingerprint is present in the sera of patients with fibrosis. An ANN fibrosis index derived from this fingerprint could differentiate between different stages of fibrosis and predict fibrosis and cirrhosis in CHB infection.

Blood Proteins↗

PubMed1997

Patient-specific predictions of outcomes in myocardial infarction for real-time emergency use: a thrombolytic predictive instrument.

BACKGROUND: Thrombolytic therapy can be life-saving in patients with acute myocardial infarction. However, if given too late or insufficiently selectively, it may provide little benefit but still cause serious complications and incur substantial costs. OBJECTIVE: To develop a thrombolytic predictive instrument for real-time use in emergency medical service settings that could 1) identify patients likely to benefit from thrombolysis and 2) facilitate the earliest possible use of this therapy. DESIGN: Creation and validation of logistic regression-based predictive instruments based on secondary analysis of clinical data. PATIENTS: 4911 patients who had acute myocardial infarction and ST-segment elevation on electrocardiogram; 3483 received thrombolytic therapy. MEASUREMENTS: Data were obtained from 13 major clinical trials and registries and directly from medical records, including electrocardiograms obtained at presentation. Input variables include presenting clinical and electrocardiography features; predictive models generate probabilities for acute (30-day) mortality if and if not treated with thrombolysis, 1-year mortality rates if and if not treated with thrombolysis, cardiac arrest if and if not treated with thrombolysis, thrombolysis-related intracranial hemorrhage, and thrombolysis-related major bleeding episode requiring transfusion. Together, these models constitute the thrombolytic predictive instrument. RESULTS: The predictive models generated the following mean predictions for patients in the Thrombolytic Predictive instrument Database: 30-day mortality rate, 7.1%; 1-year mortality rate, 10.9%; rate of cardiac arrest, 3.7%; rate of thrombolysis-related intracranial hemorrhage. 0.6%; and rate of other thrombolysis-related major bleeding episodes, 5.0%. They discriminated with between persons having and those not having the predicted outcome; areas under the receiver-operating characteristic (ROC) curve were between 0.77 and 0.84 for the five outcomes. Calibration between each instrument's predicted and observed served rates was excellent. Validation of the predictive instruments of 30-day and 1-year mortality, done on a separate test dataset, yielded areas under the ROC curve of 0.76 for each CONCLUSIONS: After the basic features of a clinical presentation are entered into a computerized electrocardiograph, the predictions of the thrombolytic predictive instrument can be printed on the electrocardiogram report. This decision aid may facilitate earlier and more appropriate use of thrombolytic therapy in patients with acute myocardial infarction.

Cerebral Hemorrhage↗

PubMed1997

A neural network model for predicting pancreas transplant graft outcome.

OBJECTIVE: To compare the results of a neural network versus a logistic regression model for predicting early (0-3 months) pancreas transplant graft survival or loss. RESEARCH DESIGN AND METHODS: This study was a cross-sectional, secondary analysis of demographic and clinical data from 117 simultaneous pancreas-kidney (SPK), 35 pancreas-after-kidney (PAK), and 8 pancreas-transplant-alone (PTA) patients (n = 160). The majority of patients were men (57%) and were white (90.1%), with a mean age of 39 +/- 8.09 years. Of the patients, 23 (14.4%) experienced early graft loss, which included any loss owing to technical or immunological causes, and death with a functional graft. Data were analyzed with a logistic regression model for multivariate analysis and a backpropagation neural network (BPNN) model. RESULTS: A total of 12 predictor variables were chosen from literature and transplant surgeon recommendations. A logistic model with all predictor variables included correctly classified 93.53% of cases. Model sensitivity was 35.71%; specificity was 100% (pseudo-R2 0.24). Of the predictors, history of alcohol abuse (odds ratio [OR] 32.39; 95% CI 1.67-626.89), having a PAK or PTA (OR 13.6; 95% CI 2.20-84.01), and use of a nonlocal organ procurement center (OPO) (OR 4.51; 95% CI 0.78-25.96) were most closely associated with early graft loss. The BPNN model with the same 12 predictor variables correctly predicted 92.50% of cases (R2 0.71). Model sensitivity was 68%; specificity was 96%. Of the predictors, the three variables most closely associated with graft outcome in this model were recipient/donor weight difference >50 lb, having a PAK or PTA, and use of a nonlocal OPO. CONCLUSIONS: First, the BPNN model correctly predicted 92.5% of graft outcomes versus the logistic model (93.53%). Second, the BPNN model rendered more accurate predictions (>0.70 = loss; <0.30 = survival) versus the logistic model (>0.50 = loss; <0.50 = survival). Third, the BPNN model was more sensitive (68%) than the logistic model (35.71%) to graft failures and demonstrated an almost threefold increase in explained variance (R2 = 0.71 vs. 0.24). These results suggest that the BPNN model is a more powerful tool for predicting early pancreas graft loss than traditional multivariate statistical models.

Adolescent↗

PubMed2004

On-line monitoring of the density of linear low-density polyethylene in a real plant by near-infrared spectroscopy and chemometrics.

This paper reports on-line monitoring of the density of linear low-density polyethylene (LLDPE) by near-infrared (NIR) spectroscopy and chemometrics. The on-line monitoring was carried out not only in a laboratory but also in a real plant. We composed an on-line monitoring system for molten polymers consisting of a Fourier transform near-infrared (FT-NIR) spectrometer, input/output (I/O) module, a personal computer, and a sampling cell that we developed. We first compared NIR spectra of LLDPE in the solid and melt states and then developed calibration models that predict the density using partial least squares regression (PLS). The sample sets for developing prediction models were collected for three months at the plant, and the density of LLDPE was continuously monitored on-line for another three months using the model. The standard error of prediction (SEP) for the on-line monitoring of the density of LLDPE at the plant was +/-2.1 mg/cm(3) (range: 0.91-0.95 g/cm(3)).

Chemical Industry↗

PubMed2006

Revisiting the hexane-water interface via molecular dynamics simulations using nonadditive alkane-water potentials.

We present results addressing properties of a polarizable force field for hexane based on the fluctuating charge (FQ) formalism and developed in conjunction with the Chemistry at Harvard Molecular Mechanics (CHARMM) potential function. Properties of bulk neat hexane, its liquid-vapor interface, and its interface with a polarizable water model (TIP4P-FQ) are discussed. The FQ model is compared to a recently modified alkane model, C27r, also based on the CHARMM potential energy function. With respect to bulk properties, both models predict bulk density within 1%; the FQ model predicts the liquid vaporization enthalpy within 2%, while the C27r force field underestimates the property by roughly 20% (and in this sense reflects the quality of the C27r force field across the spectrum of linear and branched alkanes). The FQ hexane model realistically captures the dielectric properties of the bulk in terms of a dielectric constant of 1.94, in excellent agreement with experimental values in the range of 1.9-2.02. This behavior is also in conformity with a recent polarizable alkane model based on Drude oscillators. Furthermore, the bulk dielectric is essentially captured in the infinite frequency, or optical, dielectric contribution. The FQ model is in this respect a more realistic force field for modeling lipid bilayer interiors for which most current state-of-the-art force fields do not accurately capture the dielectric environment. The molecular polarizability of the FQ model is 11.79 A3, in good agreement with the range of experimental and ab initio values. In contrast to FQ models of polar solvents such as alcohols and water, there was no need to scale gas-phase polarizabilities in order to avoid polarization catastrophes in the pure bulk. In terms of the liquid-vapor and liquid-liquid interfaces, the FQ model displays a rich orientational structure of alkane and water in the respective interfacial systems, in general conforming with earlier simulation studies of such interfaces. The FQ force field shows a marked deviation in the interfacial dipole potentials computed from the charge densities averaged over simulation trajectories. At the liquid-vapor interface, the FQ model predicts a potential drop of -178.71 mV in contrast to the C27r estimate of -433.80 mV. For the hexane-water interface, the FQ force field predicts a dipole potential drop of -379.40 mV in contrast to the C27r value of -105.42 mV. Although the surface dipole potential predicted by the FQ model is roughly 3.5 times that predicted by the C27r potential, it is consistent with reported experimental potentials across solvated lipid bilayers in the range of 400-600 mV.

Alkanes↗

PubMed2002

Structured model to predict intracellular amino acid shortages during recombinant protein overexpression in E. coli.

Recombinant protein overexpression and the classical stringent response have been shown to induce the same proteases. Since the stringent response was the result of an intracellular amino acid shortage, it was hypothesized that the overexpression of the recombinant protein also caused an intracellular amino acid shortage. A structured non-segregated kinetic mathematical model of recombinant Escherichia coli was developed to predict intracellular amino acid shortages during recombinant protein overexpression, and thus the induction of the stringent response. Two model recombinant proteins were examined, chloramphenicol acetyl-transferase (CAT) and an 'average protein'. The model predicted an aromatic amino acid shortage during CAT overexpression, as predicted based on the CAT's amino acid content. The model also predicted a shortage of the intracellular alanine family amino acid pool during CAT overexpression. This was unexpected due to the relatively low content of alanine family amino acids in CAT compared to the average E. coli protein. The model predicted alanine, glutarate, and aspartate family amino acid shortages during recombinant 'average protein' overexpression. Additionally, the model predicted a decrease in the ribosome pool at induction for both recombinant proteins, which agrees with published experimental results. Thus, the structured kinetic model was able to predict amino acid shortages, that could potentially cause a stringent response and elevated protease activity.

Amino Acids↗

PubMed2007

Early prediction of preterm birth for singleton, twin, and triplet pregnancies.

OBJECTIVES: To create prediction models of early preterm birth for singletons, twin, and triplet pregnancies. STUDY DESIGN: We used a historical cohort study with the 1996 birth registration data for singletons and the 1995-1997 linked birth/infant death dataset for multiple births of the United States. Preterm birth was defined as gestational age <32 completed weeks. Eligible study subjects were randomly allocated to two groups: one group (80% subjects) for the creation of the prediction models, and the other group (20% subjects) for the validation of the established prediction models. Multivariate logistic regressions were used to establish the prediction models. We further assessed the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of the established prediction models with different cut-off values in the validation group. RESULTS: The sensitivity, specificity, PPV, and NPV of the established model were 24.58, 93.54, 5.91, and 98.69%, respectively for singletons, 64.66, 57.04, 16.29, and 92.59%, respectively for twins, and 63.57, 53.58, 42.96, and 72.78%, respectively for triplets. CONCLUSION: The prediction models of early preterm birth for singleton, twin, and triplet pregnancies created by this study could be useful for obstetricians to identify women being at high risk of preterm birth at early gestation.

Adult↗

PubMed2005

Predictive survival model with time-dependent prognostic factors: development of computer-aided SAS Macro program.

AIMS AND OBJECTIVES: Computer program for the prediction of survival with respect to time-dependent proportional hazards regression model has been rarely addressed. We therefore developed a SAS Macro program for time-dependent Cox regression predictive model for empirical survival data associated with time-dependent covariates. METHOD: Time-dependent proportional hazards regression model and partial likelihood in association with time-varying predictors were explicitly delineated. Baseline hazard using Andersen's method was incorporated into proportional hazards regression model to predict the dynamic change of cumulative survival in respect of time-varying predictors. Two SAS Macro programs for time-dependent predictive survival model and model validation using receiver operative characteristics were written with SAS IML language. RESULTS: The computer program was applied to data on clinical surveillance of small hepatocellular carcinoma (HCC) treated by percutaneous ethanol injection (PEI) or transcatheter arterial embolization (TAE) with time-varying predictors such as alpha-feto protein (AFP) and other biological markers. CONCLUSION: The program is very useful for real-time prediction of cumulative survival on the basis of time-dependent covariates.

Carcinoma, Hepatocellular↗

PubMed2003

External validation is necessary in prediction research: a clinical example.

BACKGROUND AND OBJECTIVES: Prediction models tend to perform better on data on which the model was constructed than on new data. This difference in performance is an indication of the optimism in the apparent performance in the derivation set. For internal model validation, bootstrapping methods are recommended to provide bias-corrected estimates of model performance. Results are often accepted without sufficient regard to the importance of external validation. This report illustrates the limitations of internal validation to determine generalizability of a diagnostic prediction model to future settings. METHODS: A prediction model for the presence of serious bacterial infections in children with fever without source was derived and validated internally using bootstrap resampling techniques. Subsequently, the model was validated externally. RESULTS: In the derivation set (n=376), nine predictors were identified. The apparent area under the receiver operating characteristic curve (95% confidence interval) of the model was 0.83 (0.78-0.87) and 0.76 (0.67-0.85) after bootstrap correction. In the validation set (n=179) the performance was 0.57 (0.47-0.67). CONCLUSION: For relatively small data sets, internal validation of prediction models by bootstrap techniques may not be sufficient and indicative for the model's performance in future patients. External validation is essential before implementing prediction models in clinical practice.

Bacterial Infections↗