Selection of thrombolytic therapy for individual patients: development of a clinical model. GUSTO-I Investigators

We developed a logistic regression model with data from the GUSTO-I trial to predict mortality rate differences in individual patients who received accelerated tissue plasminogen activator (TPA) versus streptokinase treatment for acute myocardial infarction. A nomogram was developed from a reduced version of this model that approximated the underlying risk of patients treated with streptokinase, and thus the benefit of TPA. The 30-day mortality rate with accelerated TPA was 0.063 versus 0.073 with streptokinase and subcutaneously administered heparin and 0.074 with streptokinase and intravenously administered heparin. No baseline patient characteristics were significantly associated with a different relative effect of TPA. Older patients and those with anterior infarction, higher Killip classification (except Killip class IV), lower blood pressure, and increased heart rate had the greatest absolute benefit with accelerated TPA. Patients with acute myocardial infarction who had more high-risk characteristics derived a greater absolute benefit from treatment with accelerated TPA versus streptokinase.

Incidence and impact on outcome of streptokinase allergy in the GUSTO-I trial. Global Utilization of Streptokinase and t-PA in Occluded Coronary Arteries

We evaluated allergic reactions in 20,201 patients randomized to the streptokinase arms of The Global Utilization of Streptokinase and t-PA (Tissue Plasminogen Activator) in Occluded Coronary Arteries (GUSTO-I) trial, and tested the hypothesis that patients with streptokinase allergy would exhibit higher mortality. After adjusting for baseline variables and time of death, we found comparable coronary patency, left ventricular function, mortality, and bleeding complications between patients with versus those without streptokinase allergy.

Predictors of outcome of reperfusion therapy

The physician wishing to estimate the risk of death for a patient experiencing acute myocardial infarction can rapidly use parameters available on first seeing the patient to estimate the expected mortality and can update this risk during the early phase of the event. From the initial evaluation, factors that exclude patients from reperfusion therapy constitute a substantially elevated risk. Older age, haemodynamic distress, and markers of ongoing ischaemia provide a large amount of prognostic information from the initial exam. Careful surveillance for additional complications allows continuous risk assessment during the early course.

Effects of dopamine agonists on Cebus apella monkeys with previous long-term exposure to fluphenazine

Sixty-one weeks after 48 weeks of treatment with fluphenazine decanoate or placebo, 37 socially living Cebus apella monkeys were evaluated for differences in dopaminergic sensitivity by exposure to 0.75 mg/kg, i.m. of amphetamine (AMPH) (indirect agonist) and apomorphine (APOM) (direct agonist). The fluphenazine-treated animals differed (p < or = 0.05) from control animals on some hourly measures of composite behavioral variables (CBVs). Animals exposed to fluphenazine showed a greater decrease in the aggressiveness CBV and a smaller decrease in self- and environment-directed behaviors than placebo animals. CBVs for normal locomotion and directs affiliation showed no significant differences. The fluphenazine-treated group showed greater agonist induction of stereotypic behavior (p < or = 0.01), and larger decreases in prolactin response to AMPH (p < or = 0.05). Our findings indicate that following extended treatment with an antipsychotic there is increased sensitivity to dopamine, as evidenced by stereotypies and possibly hypophyseal responsiveness.

Coronary revascularization surgery after myocardial infarction: impact of bypass surgery on survival after thrombolysis. GUSTO Investigators. Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries

OBJECTIVES

This study sought to investigate the impact of surgical revascularization on outcome after myocardial infarction.

BACKGROUND

Small variations in rates of coronary artery bypass graft surgery (CABG) were noted among thrombolytic regimens in the Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries (GUSTO) trial, prompting the question of whether survival differences were partly related to differences in CABG rates.

METHODS

Patients in the GUSTO trial were randomized to one of four thrombolytic strategies. Of 40,861 patients with complete data, 3,526 underwent surgical revascularization during their initial hospital admission. Thirty-day and 1-year mortality rates were estimated using Kaplan-Meier techniques, and the impact of CABG as a time-dependent covariate on death was evaluated using a Cox survival model, adjusting for baseline prognostic factors.

RESULTS

The median time from study enrollment to CABG was 7 days across treatment groups. A 15% reduction in mortality for the tissue-type plasminogen activator (t-PA)-treated group was evident by the seventh day. Bypass surgery was a significant independent predictor of 30-day mortality (risk ratio 1.87) and a weaker predictor of 1-year mortality (risk ratio 1.21). Operative mortality was highest in patients with acute mitral regurgitation, ventricular septal defect or poor left ventricular function and in those undergoing CABG within the first 4 days of randomization.

CONCLUSIONS

The survival benefit of accelerated t-PA was not related to surgical revascularization. Bypass surgery was associated with excess mortality in the first year, but the added short-term mortality associated with CABG may be balanced by anticipated long-term benefit in specific groups of patients.

Initial clinical experience with an implantable human atrial defibrillator

Low energy biatrial shock is an effective means of restoring sinus rhythm in patients with atrial fibrillation (AF). Ventricular proarrhythmia is avoided provided that shocks are well synchronized to R waves that are not at closely coupled intervals or preceded by long-short cycles. Based on these principles, an implantable atrial defibrillator has been developed and was implanted in three patients with drug refractory paroxysmal AF. The device detects AF via an actively fixed right atrial and a self-retaining coronary sinus defibrillating leads, and delivers 3/3 ms biphasic shocks up to 300 V synchronized to the R wave. The mean implant threshold (ED50) was 195 V (1.8 J). and minimum voltage at conversion during follow-up assessments at 1, 3, and 6 months were 260 V, 2.5 J. 250 V, 2.3 J, and 300 V, 3.0 J respectively. Detection of AF was 100% specific and shocks were 100% synchronized, although only a proportion of synchronized R waves were considered suitable for shock delivery primarily because of closely coupled cycles. Three patients had 9 spontaneous AF episodes, 8/9 (89%) successfully defibrillated by shocks of 260-300 V. Sedation was not used in 4 out of 9 (45%) episodes. Backup ventricular pacing was initiated by the device in 6 out of (67%) episodes. One patient had more frequent AF after lead placement, which subsided after a change of medication. There was no ventricular proarrhythmia. It is concluded that an implantable atrial defibrillator is a viable therapy for selected patients with paroxysmal AF. The device is capable of accurate AF detection, R wave synchronization and ventricular support pacing after successful defibrillation of AF.

Admission prediction of expected final myocardial infarct size using weighted ST-segment, Q wave, and T wave measurements

Formulas for predicting final acute myocardial infarction (MI) size from ST-segment deviation on an initial electrocardiogram were proposed by Aldrich et al. for anterior and inferior infarct locations. This study of 529 patients who did not receive thrombolytic therapy was performed to determine the effectiveness of the Aldrich formulas for predicting final QRS MI size; to propose new formulas for predicting final MI size using ST-segment deviation, Q wave, and T wave information in a development population of 322 patients; and to evaluate the new formulas in a randomly selected population of 207 patients. The Aldrich formulas achieved correlations with final infarct size of r = .40 for anterior and r = .43 for inferior MI locations in the present population which are weaker than those previously reported. Formulas that consider electrocardiographic parameters in addition to ST-segment deviation were proposed for both anterior and inferior final MI size. In the test set of 207 patients, these models explained 16.9% and 15.2% of the variation in final MI size for anterior and inferior locations respectively. They may prove useful in assessing the extent of myocardial salvage where interventions are to be tested.

Age and outcome with contemporary thrombolytic therapy. Results from the GUSTO-I trial. Global Utilization of Streptokinase and TPA for Occluded coronary arteries trial

BACKGROUND

Elderly patients with acute myocardial infarction have much to gain from reperfusion with thrombolytic therapy but are also at increased risk of adverse events. We examined outcomes according to age of patients receiving thrombolysis in an international trial.

METHODS AND RESULTS

Patients were randomized to streptokinase plus subcutaneous heparin, streptokinase plus intravenous heparin, accelerated tissue plasminogen activator (TPA) plus intravenous heparin, or streptokinase and TPA plus intravenous heparin. Clinical outcomes at 30 days (death, stroke, and nonfatal, disabling stroke) and 1-year mortality were summarized descriptively for patients aged < 65 (n = 24,708), 65 to 74 (n = 11,201), 75 to 85 (n = 4625), and > 85 years (n = 412) and assessed as continuous functions of age. Older patients had a higher-risk profile with regard to baseline clinical and angiographic characteristics. Mortality at 30 days increased markedly with age (3.0%, 9.5%, 19.6%, and 30.3% in the four groups, respectively), as did stroke, cardiogenic shock, bleeding, and reinfarction. Combined death or disabling stroke occurred less often with accelerated TPA in all but the oldest patients, who showed a weak trend toward a lower incidence with streptokinase plus subcutaneous heparin: odds ratio 1.13; 95% confidence interval 0.6, 2.1. Similarly, accelerated TPA treatment resulted in lower 1-year mortality in all but the oldest patients (47% TPA versus 40.3% streptokinase).

CONCLUSIONS

Lower mortality and greater net clinical benefit were seen with accelerated TPA in patients aged < or = 85 years. Because data are limited for patients aged > 85 years, the relative superiority of a given thrombolytic regimen cannot be determined. The interactions of stroke and mortality with newer thrombolytic strategies must be examined explicitly in older patients.

One-year results from the Global Utilization of Streptokinase and TPA for Occluded Coronary Arteries (GUSTO-I) trial. GUSTO-I Investigators

BACKGROUND

In the randomized Global Utilization of t-PA and Streptokinase for Occluded Coronary Arteries (GUSTO-I) trial, 41021 patients received one of four thrombolytic regimens. Patients treated with accelerated tissue plasminogen activator (TPA) had a lower 30-day mortality rate (6.3%) than those treated with the other regimens (7.3%, combined streptokinase groups).

METHODS AND RESULTS

Each patient who was alive at 30 days was sent a return postcard to ascertain vital status at 1 year. If the postcard was not returned, the patient (or an alternate specified at randomization) was contacted by telephone. A locator service was used in the United States for patients who could not be located by these methods. Final follow-up was 96% worldwide. One-year mortality rates remained in favor of accelerated TPA (9.1%) over streptokinase with subcutaneous heparin (10.1%, P = .011) and streptokinase with intravenous heparin (10.1%, P = .009). Combination therapy had an intermediate 1-year mortality (9.9%); this outcome was statistically indistinguishable from that with streptokinase (P = .47) but was marginally different from that with accelerated TPA (P = .05).

CONCLUSIONS

The 1-year results demonstrated a saving of 10 lives per 1000 patients treated with accelerated TPA versus streptokinase and subcutaneous or intravenous heparin. Combination thrombolytic therapy had an intermediate benefit but offered no advantage over accelerated TPA treatment alone.

Economic assessment of platelet glycoprotein IIb/IIIa inhibition for prevention of ischemic complications of high-risk coronary angioplasty. EPIC Investigators

BACKGROUND

In the EPIC trial, c7E3 Fab, an antiplatelet IIb/ IIIa receptor antibody, reduced 30-day ischemic end points after high-risk coronary angioplasty by 35% and 6-month ischemic events by 23% but increased in-hospital bleeding episodes.

METHODS AND RESULTS

Of the 2099 patients randomized in EPIC, data were collected on 2038 (97%) for prospective hospital cost and major resources. Physician fees were estimated from the Medicare Fee Schedule. Regression analysis was used to examine the economic tradeoff between reduced ischemic events and increased major bleeding during the initial hospitalization. A potential cost savings of $622 per patient during the initial hospitalization from reduced acute ischemic events with c7E3 Fab was offset by an equivalent rise ($521) in costs as the result of an increase in bleeding episodes. Baseline medical costs for the bolus and infusion c7E3 Fab arm averaged $13,577 (exclusive of drug cost) compared with $13,434 for placebo (P = .42). During the 6-month follow-up, c7E3 Fab decreased repeat hospitalization rates by 23% (P = .004) and repeat revascularization by 22% (P = .04), producing a mean $1270 savings per patient (exclusive of drug cost) (P = .018). With a cost of $1407 for the bolus and infusion c7E3 Fab regimen, the cumulative net 6-month cost to switch from standard care to routine c7E3 Fab averaged $293 per patient.

CONCLUSIONS

In high-risk coronary angioplasty, aggressive platelet inhibition with c7E3 Fab, by significantly reducing ischemic events and repeat revascularization, recoups most of the cost of therapy and has the potential to pay for itself.

Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors

Multivariable regression models are powerful tools that are used frequently in studies of clinical outcomes. These models can use a mixture of categorical and continuous variables and can handle partially observed (censored) responses. However, uncritical application of modelling techniques can result in models that poorly fit the dataset at hand, or, even more likely, inaccurately predict outcomes on new subjects. One must know how to measure qualities of a model's fit in order to avoid poorly fitted or overfitted models. Measurement of predictive accuracy can be difficult for survival time data in the presence of censoring. We discuss an easily interpretable index of predictive discrimination as well as methods for assessing calibration of predicted survival probabilities. Both types of predictive accuracy should be unbiasedly validated using bootstrapping or cross-validation, before using predictions in a new data series. We discuss some of the hazards of poorly fitted and overfitted regression models and present one modelling strategy that avoids many of the problems discussed. The methods described are applicable to all regression models, but are particularly needed for binary, ordinal, and time-to-event outcomes. Methods are illustrated with a survival analysis in prostate cancer using Cox regression.

Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors

Multivariable regression models are powerful tools that are used frequently in studies of clinical outcomes. These models can use a mixture of categorical and continuous variables and can handle partially observed (censored) responses. However, uncritical application of modelling techniques can result in models that poorly fit the dataset at hand, or, even more likely, inaccurately predict outcomes on new subjects. One must know how to measure qualities of a model's fit in order to avoid poorly fitted or overfitted models. Measurement of predictive accuracy can be difficult for survival time data in the presence of censoring. We discuss an easily interpretable index of predictive discrimination as well as methods for assessing calibration of predicted survival probabilities. Both types of predictive accuracy should be unbiasedly validated using bootstrapping or cross-validation, before using predictions in a new data series. We discuss some of the hazards of poorly fitted and overfitted regression models and present one modelling strategy that avoids many of the problems discussed. The methods described are applicable to all regression models, but are particularly needed for binary, ordinal, and time-to-event outcomes. Methods are illustrated with a survival analysis in prostate cancer using Cox regression.

Hepatitis related to the Chinese medicine Shou-wu-pian manufactured from Polygonum multiflorum

Hepatitis developed in a 31-y-old pregnant Chinese woman after consumption of Shou-Wu-Pian, a proprietary Chinese medicine prepared from Polygonum multiflorum. Tests for viral hepatitis were negative and there was no evidence of other systemic disease. The herbal preparation is commonly available in the Orient and China towns in western countries. Cases of herb-induced hepatitis reported in China are reviewed.

Nonsustained ventricular tachycardia in coronary artery disease: relation to inducible sustained ventricular tachycardia. MUSTT Investigators

BACKGROUND

Many physicians believe that electrocardiographic characteristics of nonsustained ventricular tachycardia correlate with the risk for sudden death in survivors of myocardial infarction. Sustained ventricular tachycardia induced by programmed electrical stimulation has also been shown to predict sudden death.

OBJECTIVE

To determine whether electrocardiographic characteristics of spontaneous nonsustained ventricular tachycardia can predict the inducibility of sustained ventricular tachycardia by programmed electrical stimulation in patients with coronary artery disease having abnormal ventricular function.

DESIGN

Observational cohort study.

SETTING

70 clinical electrophysiology laboratories in the United States and Canada.

PATIENTS

1480 consecutive patients with coronary artery disease, left ventricular ejection fraction of 0.40 or less, and asymptomatic nonsustained ventricular tachycardia.

INTERVENTION

Electrophysiologic study attempting to induce sustained monomorphic ventricular tachycardia.

MEASUREMENTS

Daily frequency, duration, and cycle length of spontaneous episodes of nonsustained ventricular tachycardia, measured by standard electrocardiographic recordings.

RESULTS

No statistically significant difference in the frequency or duration of spontaneous nonsustained ventricular tachycardia was seen between patients with and those without inducible sustained ventricular tachycardia. Rates of spontaneous tachycardia were slightly slower in patients with inducible ventricular tachycardia than in patients without inducible ventricular tachycardia (P = 0.047), but the difference was not clinically significant.

CONCLUSION

Electrocardiographic characteristics of spontaneous nonsustained ventricular tachycardia do not predict which patients with coronary artery disease will have inducible sustained ventricular tachycardia.

Time from symptom onset to treatment and outcomes after thrombolytic therapy. GUSTO-1 Investigators

OBJECTIVES

This study sought to examine the relations among patient characteristics, time to thrombolysis and outcomes in the international GUSTO-I trial.

BACKGROUND

Studies have shown better left ventricular function and decreased infarct size as well as increased survival with earlier thrombolysis, but the relative benefits of various thrombolytic agents with earlier administration are uncertain.

METHODS

We evaluated the relations of baseline characteristics to three prospectively defined time variables: symptom onset to treatment, symptom onset to hospital arrival (presentation delay) and hospital arrival to treatment (treatment delay). We also examined the relations of delays to clinical outcomes and to the relative 30-day mortality benefit with accelerated tissue-type plasminogen activator (t-PA) versus streptokinase.

RESULTS

Female, elderly, diabetic and hypertensive patients had longer delays at all stages. Previous infarction or bypass surgery was an additional risk factor for treatment delay. Early thrombolysis was associated with lower overall mortality rate (< 2 h, 5.5%; > 4 h, 9.0%), but no additional relative benefit resulted from earlier treatment with accelerated t-PA versus streptokinase (p = 0.38). Longer presentation and treatment delays were both associated with increased mortality rate (presentation delay < 1 h, 5.6% and > 4 h, 8.6%; treatment delay < 1 h, 5.4%, and > 90 min, 8.1%). As time to treatment increased, the incidence of recurrent ischemia or reinfarction decreased, but the rates of shock, heart failure and stroke increased.

CONCLUSIONS

Earlier treatment resulted in better outcomes, regardless of thrombolytic strategy. Elderly, female and diabetic patients were treated later, adding to their already substantial risk.

Tuning the equilibrium ion affinity and selectivity of the EF-hand calcium binding motif: substitutions at the gateway position

The ion binding parameters of the EF-hand Ca2+ binding motif are carefully tuned for different biological applications. The present study examines the contribution of the ninth position of the Ca2+-coordinating EF-loop to the tuning of Ca2+ affinity and selectivity, using the model EF-loop of the Escherichia coli galactose binding protein. Eight side chains, representing the entire set of side chains commonly observed in natural EF-loop sequences, are tested at the ninth position of the model EF-loop to determine their effects on equilibrium ion binding parameters. Using the spherical metal ions of groups Ia, IIa, and IIIa and the lanthanides as probes, both the Ca2+ affinities and ionic selectivities of the engineered sites are quantitated. Neutral side chains of different size at the ninth EF-loop position [Gln (wild type), Asn, Thr, Ser, Ala, Gly] are observed to yield similar Ca2+ affinities and retain the native ability to exclude the physiological competing metal cations Na+, K+, and Mg2+. Acidic gateway side chains (Glu, Asp) are found to reduce Ca2+ affinity and shift the ionic charge selectivity as much as 10(3)-fold toward trivalent cations. Relative to the native Gln, all engineered side chains cause a partial loss of ionic size selectivity, stemming from enhanced affinities for nonphysiological large ions. Overall, the results have implications for the molecular mechanisms used by the EF-loop to control both (i) charge selectivity, which is proposed to stem from the electrostatic repulsion between the coordinating oxygens, and (ii) size selectivity, which is theorized to involve complex interactions between multiple coordinating side chains. Finally, it has recently been shown that the ninth EF-loop position serves as a "gateway" to modulate the kinetics of Tb3+ binding and release without shifting the equilibrium affinity of this ion [Drake, S. K., & Falke, J. J. (1996) Biochemistry 35, 1753-1760]. The present results confirm that isoelectric substitutions at the gateway position have little effect on Ca2+ affinity, thereby supporting the hypothesis that the gateway side chain provides kinetic tuning of Ca2+ signaling proteins independently of their Ca2+ activation thresholds.

Spectrum of electrophysiologic and electropharmacologic characteristics of verapamil-sensitive ventricular tachycardia in patients without structural heart disease

Verapamil-sensitive ventricular tachycardia (VT) is a well-recognized clinical entity that some authorities believe may result from triggered activity. Despite its uniform response to verapamil, however, there is evidence that this uncommon form of VT may not be as homogeneous as first believed. Standard intracardiac electrophysiologic techniques were used to study verapamil-sensitive VT in 32 patients (aged 38 years +/- 20 years) without evidence of structural heart disease. More than half of these patients (69%) exhibited VT with a right bundle branch block-type QRS pattern, with the remainder (31%) displaying VT with a left bundle branch block pattern. In 31% of the patients the VT could be induced by fixed-cycle length atrial pacing, whereas in 59% of patients fixed-cycle length ventricular pacing was necessary. A critical range of cycle lengths for VT induction was required in 66% of the patients. Ventricular tachycardia was initiated with single atrial premature extrastimuli in 16% of patients, single ventricular extrastimuli in 50% of patients, and double ventricular premature extrastimuli in 9% of patients. Ventricular tachycardia displaying cycle-length alternans was observed in 28% of patients. In only 19% of patients was it possible to entrain VT during pacing from the right ventricular apex. Isoproterenol infusion was required for tachycardia induction in 50% of patients, 44% of whom had VT with a left bundle branch block QRS pattern, with the remaining 56% exhibiting VT with a right bundle branch block pattern. Beta-adrenergic blockers suppressed 53% of verapamil-sensitive VT in patients tested, whereas adenosine terminated VT in 50% of patients, with 81% of these patients exhibiting either a left bundle branch block QRS pattern or isoproterenol dependence. Ventricular tachycardia exhibiting a left bundle branch block pattern was more likely to be isoproterenol dependent (p <0.05) and adenosine sensitive (p <0.001). However, verapamil-sensitive, catecholamine-dependent VT was no more likely to be adenosine sensitive than the catecholamine-independent form of the arrhythmia (p >0.5). Verapamil-sensitive VT exhibits properties expected of both a reentrant and triggered arrhythmia, and it is inconsistently dependent on both exogenous catecholamines for induction and intravenous adenosine for termination. Verapamil-sensitive VT encompasses a heterogeneous group of tachycardias that may result from multiple cellular electrophysiologic mechanisms.

Activated partial thromboplastin time and outcome after thrombolytic therapy for acute myocardial infarction: results from the GUSTO-I trial

BACKGROUND

Although intravenous heparin is commonly used after thrombolytic therapy, few reports have addressed the relationship between the degree of anticoagulation and clinical outcomes. We examined the activated partial thromboplastin time (aPTT) in 29,656 patients in the Global Utilization of Streptokinase and t-PA for Occluded Coronary Arteries (GUSTO-I) trial and analyzed the relationship between the aPTT and both baseline patient characteristics and clinical outcomes.

METHODS AND RESULTS

Intravenous heparin was administered as a 5000-U bolus followed by an initial infusion of 1000 U/h, with dose adjustment to achieve a target aPTT of 60 to 85 seconds. aPTTs were collected 6, 12, and 24 hours after thrombolytic administration. Higher aPTT at 24 hours was strongly related to lower patient weight (P < .00001) as well as older age, female sex, and lack of cigarette smoking (all PT< .0001). At 12 hours, the aPTT associated with the lowest 30-day mortality, stroke, and bleeding rates was 50 to 70 seconds. There was an unexpected direct relationship between the aPTT and the risk of subsequent reinfarction. There was a clustering of reinfarction in the first 10 hours after discontinuation of intravenous heparin.

CONCLUSIONS

Although the relationship between aPTT and clinical outcome was confounded to some degree by the influence of baseline prognostic characteristics, aPTTs higher than 70 seconds were found to be associated with higher likelihood of mortality, stroke, bleeding, and reinfarction. These findings suggest that until proven otherwise, we should consider the aPTT range of 50 to 70 seconds as optimal with intravenous heparin after thrombolytic therapy.

Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors

Multivariable regression models are powerful tools that are used frequently in studies of clinical outcomes. These models can use a mixture of categorical and continuous variables and can handle partially observed (censored) responses. However, uncritical application of modelling techniques can result in models that poorly fit the dataset at hand, or, even more likely, inaccurately predict outcomes on new subjects. One must know how to measure qualities of a model's fit in order to avoid poorly fitted or overfitted models. Measurement of predictive accuracy can be difficult for survival time data in the presence of censoring. We discuss an easily interpretable index of predictive discrimination as well as methods for assessing calibration of predicted survival probabilities. Both types of predictive accuracy should be unbiasedly validated using bootstrapping or cross-validation, before using predictions in a new data series. We discuss some of the hazards of poorly fitted and overfitted regression models and present one modelling strategy that avoids many of the problems discussed. The methods described are applicable to all regression models, but are particularly needed for binary, ordinal, and time-to-event outcomes. Methods are illustrated with a survival analysis in prostate cancer using Cox regression.

Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors

Multivariable regression models are powerful tools that are used frequently in studies of clinical outcomes. These models can use a mixture of categorical and continuous variables and can handle partially observed (censored) responses. However, uncritical application of modelling techniques can result in models that poorly fit the dataset at hand, or, even more likely, inaccurately predict outcomes on new subjects. One must know how to measure qualities of a model's fit in order to avoid poorly fitted or overfitted models. Measurement of predictive accuracy can be difficult for survival time data in the presence of censoring. We discuss an easily interpretable index of predictive discrimination as well as methods for assessing calibration of predicted survival probabilities. Both types of predictive accuracy should be unbiasedly validated using bootstrapping or cross-validation, before using predictions in a new data series. We discuss some of the hazards of poorly fitted and overfitted regression models and present one modelling strategy that avoids many of the problems discussed. The methods described are applicable to all regression models, but are particularly needed for binary, ordinal, and time-to-event outcomes. Methods are illustrated with a survival analysis in prostate cancer using Cox regression.

The gene for the human architectural transcription factor HMGI-C consists of five exons each coding for a distinct functional element

The gene on chromosome 12 coding for the human protein HMGI-C has been cloned and partially sequenced. It consists of five exons, the first and last of which include long untranslated regions. The 5' UTR includes a (CA/T)n tract and a polymorphic (CT)n tract. Exons II, III and IV (87, 51 and 33 bp) are dispersed over > 30 kb. Exons I-III separately encode the three basic DNA binding domains ('A-T hooks'), exon IV encodes an 11 amino acid sequence characteristic of HMGI-C and absent from the human HMGI(Y) gene [Friedmann, M., Holth, L. T., Zoghbi, H. Y. and Reeves, R. (1993) Nucleic Acids Res., 21, 4259-4267], whilst exon V encodes the acidic C-terminal domain, which is subject to multiple phosphorylation. The HMGI-C gene is thus a striking example of the separation of functional protein elements into different coding exons.

Job strain and the prevalence and outcome of coronary artery disease

BACKGROUND

It has been hypothesized that jobs that have both high psychological demands and low decision latitude ("job strain") can lead to coronary disease. The objective of this study was to test whether job strain was correlated with the presence of coronary disease at angiography or with long-term outcome in patients with angiographic coronary disease.

METHODS AND RESULTS

Employed patients under the age of 65 years undergoing diagnostic coronary angiography completed a self-administered questionnaire about their job duties and work environment. Job strain was measured by the method of Karasek. Patients were separated into three groups, based on extent of coronary disease: significant disease (> or = 75% stenosis), insignificant disease (> 0% but < 75% stenosis), and normal coronary arteries. Statistical analyses were performed using logistic regression and the Cox proportional hazards model. The 1489 patients enrolled had a median age of 52 years; 76% were male and 88% were white. By design, all patients were employed, 60% in white-collar jobs and only 16% in jobs requiring heavy labor. Traditional cardiac risk factors were most prevalent in the 922 patients with significant coronary artery disease, at intermediate levels in the 204 patients with insignificant disease, and least prevalent in the 363 patients with normal coronary arteries (all P < .01). Job strain was actually more common in patients with normal coronary arteries (35%) than in patients with insignificant (26%) or significant disease (25%, P < .002). In a multivariate analysis, job strain was not significantly correlated with the presence of coronary disease. Job strain was not correlated with angina frequency at the time of angiography. Job strain was not a predictor of cardiac events (cardiac death or nonfatal myocardial infarction) during follow-up.

CONCLUSIONS

Job strain was not correlated with the prevalence or severity of coronary artery disease in a cohort of patients undergoing coronary angiography. The outcome of patients with angiographically defined coronary disease was not affected by the level of job strain as measured by the method of Karasek.