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 1,459 records · Page 81Linked to original sources

PubMed2005

Multiple imputation of the Glasgow Coma Score.

BACKGROUND: To investigate whether multiple imputation (MI) of missing Glasgow Coma Scale (GCS) values generates more accurate GCS frequency distributions and less biased parameter estimates in logistic regression models predicting mortality than the standard procedure of excluding observations with missing GCS values. METHODS: The study population consisted of 5,065 patients with complete GCS information from the trauma registry of a Level 1 trauma center. Missing GCS values were imposed on the data set, and the performance of MI (extrapolating missing GCS from a data prediction model) and of deleting all data observations with missing GCS (list-wise deletion) were evaluated. GCS and Trauma and Injury Severity Score (TRISS) frequency distributions and parameter estimates were compared with true values from the original data set. RESULTS: GCS and TRISS frequency values generated by MI were much more accurate than those generated by list-wise deletion. GCS and TRISS parameter estimates generated by MI all had acceptable bias and coverage rates when compared with true values. List-wise deletion provided biased parameter estimates for the GCS, the Revised Trauma Score, and the Injury Severity Score. CONCLUSION: MI is a valid solution to the problem of missing GCS data in trauma research. It allows the conservation of precious data observations and leads to unbiased estimates in consequent analyses. Analyses, which exclude observations with missing GCS data, provide biased results.

Biomedical Research↗

PubMed1996

Selective endoscopic retrograde cholangiography and preoperative bile duct stone removal in patients scheduled for laparoscopic cholecystectomy: a prospective study.

OBJECTIVES: Laparoscopic cholecystectomy (LC) has become the treatment of choice for patients with symptomatic cholelithiasis. About 10% of patients with symptomatic gallstones may bear associated common bile duct (CBD) stones. The preferred approach to these patients is the removal of CBD stones by endoscopic retrograde cholangiography (ERCP) before LC. However, ERCP before LC should be performed only in patients with suspected choledocholithiasis. The aims of this study were to: 1) generate an efficacious predictive model for selecting patients with suspected choledocholithiasis to submit to preoperative ERCP, and 2) test the safety of the endoscopic/laparoscopic procedure. METHODS: Historical, biochemical, and ultrasonographic data were collected prospectively. Receiver operating characteristics curve analysis was adopted for determining optimal biochemical and ultrasonographic cut-off values. Multivariate analysis using logistic regression with generation of the best model identifying independent predictors of CBD stones was also employed. RESULTS: The optimal model predicted a 95% probability of CBD stones in a patient who presented with elevated ALP (over 300 IU/L) and ALT (over 40 IU/L) levels and CBD dilation > 8 mm at ultrasonography. Endoscopic removal of CBD stones was achieved in 95% of patients, with minimal morbidity and no mortality. CONCLUSIONS: The identified independent predictors of CBD stones are highly efficient selectors of patients with choledocholithiasis. Moreover, endoscopic removal of CBD stones before LC is a safe and efficacious procedure.

Adolescent↗

PubMed2004

Assessment of operative risk in colorectal cancer surgery: the Cleveland Clinic Foundation colorectal cancer model.

INTRODUCTION: Predictive models play a pivotal role in the provision of risk-adjusted, operative mortality rates. The purpose of the study was to describe the development of a dedicated prognostic index for quantifying operative risk in colorectal cancer surgery. METHODS: Data were collected from 5,034 consecutive patients undergoing major surgery in a single center from October 1976 to July 2002. Primary end point was 30-day operative mortality. A multilevel Bayesian logistic regression model was developed to adjust for case-mix and accommodate the variability of outcomes between surgeons. The model was internally validated (split-sample) and tested using measures of discrimination, calibration, and subgroup analysis. RESULTS: The patients' median age was 66 (range, 18-98) years. Operative mortality was 2.3 percent with no significant variability between surgeons or through time. Multivariate analysis identified the following independent risk factors: age (odds ratio = 1.5 per 10-year increase), American Society of Anesthesiologists grade (odds ratio for ASA II, III, IV-V vs. I = 2.6, 4.3, 6.8), TNM staging (odds ratio for Stage IV vs. I-III = 2.6), mode of surgery (odds ratio for urgent vs. nonurgent = 2.1) no-cancer resection vs. cancer resection (odds ratio = 4.5), and hematocrit level. The model offered adequate discrimination (area under receiver operator characteristic curve = 0.801) and excellent agreement between observed and model-predicted outcomes over ten major colorectal procedures (P = 0.191). CONCLUSIONS: The colorectal cancer model provided an accurate means of estimating risk for individual patients in the preoperative setting. It has important implications in everyday practice, because it may be used as an adjunct in the process of informed consent and for monitoring surgical performance through time.

Adolescent↗

PubMed1997

Predictions of mathematical models of tissue oxygenation and generation of singlet oxygen during photodynamic therapy.

Photodynamic therapy (PDT) is a relatively new protocol for cancer treatment which has recently been approved for limited clinical use. Traditionally, the success of treatment with PDT has been compared on the basis of total light delivery. Using the mathematical model of Henning et al. (Radiat. Res. 142, 221-226, 1995), we have determined that when oxygen is not depleted from the tissue, the concentration of singlet oxygen that is generated is directly proportional to the product of the light fluence rate (phi) and the concentration of the photosensitizer (Cs). Therefore, phiCs is an appropriate parameter for comparing the potential success of PDT protocols under these conditions. For a treatment of time t, the observed photodynamic effect resulting from singlet oxygen exposure should be directly related to phiCst. For high phiCs, the model predicts that oxygen depletion occurs within the tumor tissue. As a result, the photodynamic effect is no longer proportional to phiCst. We have expanded the model of Henning et al. to include the changes in oxygen concentration which occur within the capillary as blood flows through the tissue. Our new predictions with the mathematical model for optimal PDT treatment conditions are significantly different from those predicted by the previous models. Predictions of the model are given using parameters relevant for treatment of solid tumors with Photofrin.

Dihematoporphyrin Ether↗

PubMed1996

Research to develop a predicting system of mammalian subacute toxicity, (3). Construction of a predictive toxicokinetics model.

A new predictive toxicokinetics model was developed to estimate subacute toxicity (target organs, severity, etc.) of non-congeneric industrial chemicals, where the chemical structures and physico-chemical properties are only available. Thus, a physiological pharmacokinetics model, which consists of blood, liver, kidney (these were experimentally found as major toxicological targets), muscle and fat compartments, was established to simulate the chemical concentrations in organs/tissues with pharmacokinetic parameters by means of Runge-Kutta-Gill algorithm. The pharmacokinetic parameters, i.e. absorption rate, absorption ratio, hepatic extraction ratio of metabolism and renal clearance were calculated by using separately established Quantitative Structure-Pharmacokinetics Relationship equations. The developed predictive model was then applied to simulations of 43 non-congeneric industrial chemicals. The chemical concentrations in organs/tissues after single oral administration were simulated, and their maximum concentrations (Cmax's) and area under the concentration-time curves (AUC's) were calculated. Fast Inverse Laplace Transform was newly applied for the purpose of simulation of 28-day repeated dose toxicity. Simulated concentrations of 28 days repeated dose were, however, found to be the same as those of simple repetitions of a single administration per day because of the short half-lives of non-congeneric industrial chemicals. A comparison of subacute toxicity data with Cmax's and AUC's in a single dose scenario suggested that the organs/tissues with relatively high concentrations of tested chemical substances were the most sensitive targets within a chemical. Chemical concentrations in liver, for instance, were correlated with the severity of hepatotoxicity among the chemicals. It was also suggested that to improve and widen the present approach, data of metabolite and reactivity of non-congeneric industrial chemicals to organs/tissues, receptors, etc. should be incorporated into the model.

Adipose Tissue↗

PubMed1987

Gastrointestinal absorption of americium-241 by orally exposed swine: comparison of experimental results with predictions of metabolic models.

Two groups of 11-week-old swine (40 miniature and 40 domestic swine) received a single oral administration of 1.9 X 10(8) Bq (5.2 mCi) of 241Am citrate, and groups of eight animals, four of each type, were killed and sampled at 1, 2, 4, 8, 16, 24, 48, 72, and 96 h and 30 days later. Uptake and excretion patterns of the radioactivity appeared to occur in three phases: rapid uptake, rapid excretion, and then a slower excretion. All animals were systematically dissected, and the eviscerated carcasses were autoclaved for separation of bone and muscle. The predominant site of deposition was bone, and autoclaving had little effect on releasing 241Am from either bone or muscle. The maximum fractional gastrointestinal absorption of 1.1 X 10(-3) occurred 8 h after radionuclide administration. The tissue distribution data suggest partitions of 50, 20, and 30%, for bone, liver, and other soft tissues, respectively. Two metabolic models were evaluated: a modified Mewhinney-Griffith model and the ICRP 30 model to compare the biological data with model predictions. All models underestimated the actual early time data, but the fits to the experimental results were better at later times.

Administration, Oral↗

PubMed1997

Predicting endoscopic diagnosis in the dyspeptic patient. The value of predictive score models.

BACKGROUND: Score models to predict endoscopic diagnosis in dyspepsia may compensate for the unreliable clinical diagnosis. This study aimed to construct and test score models designed to predict diagnosis in dyspepstic patients managed in primary care. METHODS: Three models to predict organic dyspepsia, major dyspepsia, or peptic ulcer were constructed by regression analysis of clinical data from 1026 consecutive dyspeptic patients referred for endoscopy. The models were tested in 207 patients in primary care, who were potential candidates for endoscopy. Validation experiments were analysed using receiver operating characteristic (ROC) curves. RESULTS: Significant losses of predictive power were found for all models when applied to primary care patients, and no model could be used as a reliable decision support instrument in primary care. CONCLUSIONS: Predictive score models developed in patients referred for endoscopy are not reliable when applied to patients in primary care who are potential candidates for endoscopy. Future models should be constructed and validated in unselected primary care populations.

Decision Support Techniques↗

PubMed2006

Development and validation of two models for early prediction of response to therapy in genotype 1 chronic hepatitis C.

Early prediction of response to therapy in genotype 1 chronic hepatitis C is difficult. Two predictive models, a pretreatment scoring model (PreT-SM) and a fourth week of therapy scoring model (4w-SM) were constructed in a cohort of 104 patients from a single center (estimation cohort) and validated in a cohort of 141 patients from four independent centers (validation cohort). Individual scores were calculated using variables independently associated with sustained virological response (SVR). Baseline viral load, aspartate aminotransferase/alanine aminotransferase ratio, serum cholesterol, and a numerical score for noninvasive estimation of liver fibrosis were included in the PreT-SM; HCV RNA clearance and PreT-SM scores were included in the 4w-SM. Receiver operating characteristic analysis revealed the area under the curve in the estimation cohort and in the validation cohort to be, respectively, 0.856 and 0.847 for the PreT-SM and 0.908 and 0.907 for the 4w-SM. Low scores were associated with SVR, high scores with non-SVR. The best cutoff scores from the PreT-SM (7 and 9.70) identified, respectively, 36% of patients with SVR and 41% of those with non-SVR from the validation cohort, with high accuracy (> or =90% positive predictive value [PPV] and specificity). Similarly, cutoff scores of 3.20 and 5.60 from the 4w-SM identified, respectively, 71% of patients with SVR and 53% of those with non-SVR from the same cohort with high accuracy (PPV and specificity >92%). In conclusion, these models predicted response to therapy before or after 4 weeks of treatment in approximately 60% of genotype 1 patients and may be valuable for the management of this condition.

Adult↗

PubMed2002

A stroke prediction score in the elderly: validation and Web-based application.

The objective of this study was to construct a prediction model for predicting stroke in an elderly U.S. population, and to assess the accuracy in this population of other previously published prediction models. The subjects were participants in the Cardiovascular Health Study: 2,495 men and 3,393 women age 65 years and older at baseline, and followed for 6.3 years. Among 5,711 participants free of baseline stroke, 399 strokes occurred. Sex-specific prediction equations were constructed using study variables that were most importantly related to incident stroke: age, systolic blood pressure, diabetes, ECG diagnosis of atrial fibrillation or left ventricular hypertrophy, confirmed history of cardiovascular disease, diabetes, time to walk 15 ft, and serum creatinine. The prediction rule was implemented as a risk score and in a Web-based interactive Java applet. Overall, the model predicted 5-year stroke risks ranging from less than 1 to 59%. The 20% of subjects in the highest predicted risk group had a 5-year actual stroke incidence rate of 15%, while the 20% lowest risk group had a 1% incidence. Risk scores from two other studies performed well in these study participants. Effective discrimination between low and high stroke risk in the elderly was possible in this cohort with data that are easy to obtain. Evaluation of the generalizability and clinical usefulness of this prediction model requires further research.

Aged↗

PubMed2004

Nonlinear prediction for Gaussian mixture image models.

Prediction is an essential operation in many image processing applications, such as object detection and image and video compression. When the images are modeled as Gaussian, the optimal predictor is linear and easy to obtain. However, image texture and clutter are often non-Gaussian, and, in such cases, optimal predictors are difficult to obtain. In this paper, we derive an optimal predictor for an important class of non-Gaussian image models, the block-based multivariate Gaussian mixture model. This predictor has a special nonlinear structure: it is a linear combination of the neighboring pixels, but the combination coefficients are also functions of the neighboring pixels, not constants. The efficacy of this predictor is demonstrated in object detection experiments where the prediction error image is used to identify "hidden" objects. Experimental results indicate that when the background texture is nonlinear, i.e., with fast-switching gray-level patches, it performs significantly better than the optimal linear predictor.

Algorithms↗

PubMed2005

Integrating seasonal climate prediction and agricultural models for insights into agricultural practice.

Interest in integrating crop simulation models with dynamic seasonal climate forecast models is expanding in response to a perceived opportunity to add value to seasonal climate forecasts for agriculture. Integrated modelling may help to address some obstacles to effective agricultural use of climate information. First, modelling can address the mismatch between farmers' needs and available operational forecasts. Probabilistic crop yield forecasts are directly relevant to farmers' livelihood decisions and, at a different scale, to early warning and market applications. Second, credible ex ante evidence of livelihood benefits, using integrated climate-crop-economic modelling in a value-of-information framework, may assist in the challenge of obtaining institutional, financial and political support; and inform targeting for greatest benefit. Third, integrated modelling can reduce the risk and learning time associated with adaptation and adoption, and related uncertainty on the part of advisors and advocates. It can provide insights to advisors, and enhance site-specific interpretation of recommendations when driven by spatial data. Model-based 'discussion support systems' contribute to learning and farmer-researcher dialogue. Integrated climate-crop modelling may play a genuine, but limited role in efforts to support climate risk management in agriculture, but only if they are used appropriately, with understanding of their capabilities and limitations, and with cautious evaluation of model predictions and of the insights that arises from model-based decision analysis.

Climate↗

PubMed2005

An integrated model for the prediction of preeclampsia using maternal factors and uterine artery Doppler velocimetry in unselected low-risk women.

OBJECTIVE: The purpose of this study was to develop a predictive model for preeclampsia. STUDY DESIGN: This was a prospective screening study for preeclampsia using uterine artery Doppler ultrasound in unselected low-risk singleton pregnancies at community hospitals in the UK (n = 32,157). Logistic regression models were developed and their predictive ability assessed using the area under the receiver operator curve (AROC). RESULTS: Six hundred twelve (2.0%) women developed preeclampsia, and 144 (0.5%) required early delivery (<34 weeks). A model using both maternal and ultrasound factors had an AROC of 0.798, which was higher than ultrasound alone (0.729, P < .0001) or maternal factors alone (0.712, P < .0001). In early onset disease, the ROC of ultrasound alone (0.922) was not significantly improved by adding maternal predictors (0.945, P = .27). In contrast, late onset disease was better predicted by the combined model (AROC 0.798) than ultrasound alone (AROC 0.729, P < .0001) or maternal factors alone (AROC 0.712, P < .0001). CONCLUSION: The combination of uterine artery Doppler ultrasound and maternal factors provided the best estimate of risk.

Adult↗

PubMed2001

A simple prediction rule and a neural network model to predict pancreatic beta-cell reserve in young adults with diabetes mellitus.

In the present study we developed and assessed the performance of a simple prediction rule and a neural network model to predict beta-cell reserve in young adults with diabetes. Eighty three young adults with diabetes were included in the study. All were less than 40 years old and without apparent secondary causes of diabetes. The subjects were randomly allocated to 2 groups; group 1 (n = 59) for developing a prediction rule and training a neural network, group 2 (n = 24) for validation purpose. The prediction rule was developed by using stepwise logistic regression. Using stepwise logistic regression and modification of the derived equation, the patient would be insulin deficient if 3(waist circumference in cm) + 4(age at diagnosis) < 340 in the absence of previous diabetic ketoacidosis (DKA) or < 400 in the presence of previous DKA. When tested in the validation set, the prediction rule had positive and negative predictive values of 86.7 per cent and 77.8 per cent respectively with 83.3 per cent accuracy while the ANN model had a positive predictive value of 88.2 per cent and a negative predictive value of 100 per cent with 91.7 per cent accuracy. When testing the performance of the prediction rule and the ANN model compared to the assessment of 23 internists in a subgroup of 9 diabetics whose age at onset was less than 30 years and without a history of DKA, the ANN had the highest ability to predict beta-cell reserve (accuracy = 88.9), followed by the prediction rule (accuracy = 77.8%) and assessments by internists (accuracy = 60.9%). We concluded that beta-cell reserve in young adults with diabetes mellitus could be predicted by a simple prediction rule or a neural network model. The prediction rule and the neural network model can be helpful clinically in patients with mixed clinical features of type 1 and type 2 diabetes.

Adolescent↗

PubMedDate not supplied

Pesticide volatilization from soil: lysimeter measurements versus predictions of European registration models.

A comparison was drawn between model predictions and experimentally determined volatilization rates to evaluate the volatilization approaches of European registration models. Volatilization rates of pesticides (14C-labeled parathion-methyl, fenpropimorph, and terbuthylazine and nonlabeled chlorpyrifos) were determined in a wind-tunnel experiment after simultaneous soil surface application on Gleyic Cambisol. Both continuous air sampling, which quantifies volatile losses of 14C-organic compounds and 14CO2 separately, and the detection of soil residues allow for a mass balance of radioactivity of the 14C-labeled pesticides. Recoveries were found to be > 94% of the applied radioactivity. The following descending order of cumulative volatilization was observed: chlorpyrifos > parathion-methyl > terbuthylazine > fenpropimorph. Due to its high air-water partitioning coefficient, nonlabeled chlorpyrifos was found to have the highest cumulative volatilization (44.4%) over the course of the experiment. Volatilization flux rates were measured up to 993 microg m(-2) h(-1) during the first hours after application. Parameterization of the Pesticide Emission Assessment at Regional and Local Scales (PEARL) model and the Pesticide Leaching Model (PELMO) was performed to mirror the experimental boundary conditions. In general, model predictions deviated markedly from measured volatilization rates and showed limitations of current volatilization models, such as the uppermost compartment thickness, making an enormous influence on predicted volatilization losses. Experimental findings revealed soil moisture to be an important factor influencing volatilization from soil, yet its influence was not reflected by the model calculations. Future versions of PEARL and PELMO ought to include improved descriptions of aerodynamic resistances and soil moisture dependent soil-air partitioning coefficients.

Environmental Monitoring↗

PubMed2000

Effects of antiviral usage on transmission dynamics of herpes simplex virus type 1 and on antiviral resistance: predictions of mathematical models.

Herpes simplex virus type 1 (HSV-1) causes recurrent herpes labialis (RHL), a common disease afflicting up to 40% of adults worldwide. Mathematical models are used to analyze the effect of antiviral treatment on the transmission of, and the prevalence of drug resistance in, HSV-1 in the United States. Three scenarios are analyzed: no antiviral use, the current level of use, and a substantial increase in nucleoside analogue use, such as might occur if topical penciclovir were available over-the-counter for the treatment of RHL. A basic model predicts that present level of nucleoside analogue use has a negligible effect on HSV-1 transmission and that even if use of topical penciclovir for (RHL) increased substantially, the overall prevalence of infectious HSV-1 is unlikely to be reduced by more than 5%. An expanded model, which allows for acquired resistance and includes immunocompromised hosts and other more realistic features, predicts that current antiviral use is unlikely to lead to any noticeable increase in resistance. If antiviral use increases, the resulting rise in resistance in the population will depend primarily on the probability that immunocompetent hosts will acquire permanent resistance upon treatment. This probability is known to be small, but its exact value remains uncertain. If acquired resistance occurs less than once per 2,500 treated episodes, then in the community at large, the frequency of HSV-1 resistance is predicted to increase slowly, if at all (remaining below 0.5% for >50 years), even with extensive nucleoside analogue use. If acquired resistance emerges in 1 of 625 treated episodes (the maximum of an approximate 95% confidence interval derived from the results of several studies of resistance in treated hosts), then the prevalence of infection with resistant HSV-1 could rise from about 0.2% to 1.5 to 3% within 50 years. The limitations of existing data on acquired resistance and the potential impact of acquired resistance if it occurs are discussed, and strategies are suggested for enhancing information on acquired resistance. The predictions of this model contrast with the more rapid increases in antimicrobial resistance anticipated by models and observed for other pathogenic bacteria and viruses. The reasons for these contrasting predictions are discussed.

Aging↗

PubMed2002

Predictive accuracy of a model of volatile anesthetic uptake.

UNLABELLED: A computer program that models anesthetic uptake and distribution has been in use in our department for 20 yr as a teaching tool. New anesthesia machines that electronically measure fresh gas flow rates and vaporizer settings allowed us to assess the performance of this model during clinical anesthesia. Gas flow, vaporizer settings, and end-tidal concentrations were collected from the anesthesia machine (Datex S/5 ADU) at 10-s intervals during 30 elective anesthetics. These were entered into the uptake model. Expired anesthetic vapor concentrations were calculated and compared with actual values as measured by the patient monitor (Datex AS/3). Sevoflurane was used in 16 patients and isoflurane in 14 patients. For all patients, the median performance error was -0.24%, the median absolute performance error was 13.7%, divergence was 2.3%/h, and wobble was 3.1%. There was no significant difference between sevoflurane and isoflurane. This model predicted expired concentrations well in these patients. These results are similar to those seen when comparing calculated and actual propofol concentrations in propofol infusion systems and meet published guidelines for the accuracy of models used in target-controlled anesthesia systems. This model may be useful for predicting responses to changes in fresh gas and vapor settings. IMPLICATIONS: We compared measured inhaled anesthetic concentrations with those predicted by a model. The method used for comparison has been used to study models of propofol administration. Our model predicts expired isoflurane and sevoflurane concentrations at least as well as common propofol models predict arterial propofol concentrations.

Adolescent↗

PubMed2002

Estimation of the Doppler ultrasound umbilical maximal waveform envelope: II. Prediction of fetal distress.

Blood flow variables obtained via Doppler ultrasound (US) waveform estimation have been investigated for prediction of fetal distress. The umbilical flow was assessed using a number of waveform summary statistics in addition to the currently used resistance indices. We examined the relationship between umbilical artery waveform patterns and intrauterine growth restriction, preterm delivery and hypertensive disorders. To enhance prediction, we defined waveform skewness profiles based on pivotal points of the umbilical waveform that appeared to be related to the incidence of preterm delivery and that facilitated construction of IUGR prediction models. The data comprised 204 unselected pregnancies with the umbilical artery images recorded at 18 pregnancy weeks. The sample was divided into 114 pregnancies used to estimate model parameters and 90 pregnancies to validate the model. Logistic prediction models for detection of abnormal velocity waveforms associated with intrauterine growth restriction were derived, based on the waveform information. The estimated model sensitivity and specificity on the training data were 74% and 84%, respectively. Validation of the model on independent data yielded a sensitivity of 57% and specificity of 84%. The logistic IUGR prediction model appears to have significant predictive ability and potential for clinical use, even at this early gestational age. Our data suggest that prediction of IUGR at 18 pregnancy weeks can be much improved when the waveform shape is captured with a number of summary statistics in addition to resistance indices.

Algorithms↗

PubMed1993

Models for predicting accidents at junctions where pedestrians and cyclists are involved. How well do they fit?

The coefficient of determination, R2, i.e. the squared correlation coefficient between observed and fitted values, is often used as a measure of how well a model predicts the number of accidents at road junctions, for instance. The purpose of this article is to show that the R2 values obtained in different studies are rarely comparable with each other and that a prediction model can be "nearly perfect" even if the coefficient of determination is small. Another purpose of the article is to present some results of interest from a practical viewpoint in regard to accidents where pedestrians and cyclists are involved. Empirical R2 values for models predicting accidents at junctions where pedestrians or cyclists are involved are compared with the maximal R2 values that could possibly be obtained. The latter can be calculated both theoretically and with the aid of simulation. How the maximal R2 value depends on the average accident level and the relative dispersion of the expected values for the studied junctions is also shown theoretically. The results obtained show how difficult it can be to determine whether and how far the number of accidents is influenced by additional factors, over and above the traffic flows, which describe the design in greater detail.

Accidents, Traffic↗