Multivariable regression models are widely used in medical literature for the purpose of diagnosis or prediction. Conventionally, the adequacy of these models is assessed using metrics of diagnostic performances such as sensitivity and specificity, which fail to account for clinical utility of a specific model. Decision curve analysis (DCA) is a widely used method to measure this utility. In this framework, a clinical judgment of the relative value of benefits (treating a true positive case) and harms (treating a false positive case) associated with prediction models is made. As such, the preferences of patients or policy-makers are accounted for by using a metric called threshold probability. A decision analytic measure called net benefit is then calculated for each possible threshold probability, which puts benefits and harms on the same scale. The article is a technical note on how to perform DCA in R environment. The decision curve is depicted with the system. Correction for overfitting is done via either bootstrap or cross-validation. Confidence interval and P values for the comparison of two models are calculated using bootstrap method. Furthermore, we describe a method for computing area under net benefit for the comparison of two models. The average deviation about the probability threshold (ADAPT), which is a more recently developed index to measure the utility of a prediction model, is also introduced in this article.

The receiver operating characteristic curve (ROC) and its associated summary index, the area under the curve (AUC), have recently found an increasingly popular place in medical diagnosis and population screening surveys. Nevertheless, the index may erroneously rate a perfect or nearly perfect marker as having no diagnostic or screening value. In this paper, we propose two new summary indices, the projected length of the ROC curve (PLC) and the area swept out by the ROC curve (ASC), to summarize the ROC curve. Like the conventional AUC, these new indices both have clear probabilistic interpretations and are easily defined geometrically. In addition, they do not suffer from the shortcoming of the AUC. These properties render them good alternatives for evaluating the overall performance of a diagnostic or screening test.

In epidemiology, the comparative mortality figure and the standardized mortality ratio are standardized measures in common use. Both are weighted averages of rate ratios (or observed/expected death count ratios) on the arithmetic scale. I propose a new standardized measure, the geometrically averaged ratio (GAR), which is defined through simple averaging on the logarithmic scale. I show that, in addition to providing a valid comparison between populations, the geometrically averaged ratio possesses the following desirable properties: (1) invertibility and invariance of standardized sex ratios and (2) interpopulational comparability with different standards.

No abstract

A retrospective study of 954 resectable gastric cancers in a single institute of Taiwan from 1971 to 1990 was performed to evaluate improvements in gastric cancer surgery. The patients were divided into four time periods representing an overall experience of progressive implementation of aggressive resection and increased extent of systematic lymph node dissection. The clinicopathologic data and survival rates were statistically compared and the significance of the extent of resection on survival analyzed. A significant increase in the proportion of upper one-third tumors (from 14.8% to 20.4%) and a decrease in the incidence of intestinal type (73.6% to 41.5%) was found within the overall period. The proportion of patients with early gastric cancer increased from 11.5% to 19.4%. Patients who underwent total gastrectomy and combined visceral resection increased from 13.7% to 27.4% and 19.8% to 41.1%, respectively. An increase of both total dissected lymph node number and the incidence of detected lymph node metastases in early gastric cancer were associated with more extensive lymphadenectomy. An improved 5-year survival rate following aggressive resection was found for all stages except stage IV and T4 lesions, and the surgical mortality decreased from 5.5% to 2.0%. Patients with earlier stage lesions benefited more from radical resection, especially those with stage II and T2 lesions. Systematic lymph node dissection increased the 5-year survival of patients by about 10% for stage III or T3 lesions but not for patients with stage IV or T4 lesions. Multivariate analysis confirmed the significance of the improved technique of lymphadenectomy on the prognosis of gastric cancer following resection in Taiwan.(ABSTRACT TRUNCATED AT 250 WORDS)

No abstract

Study objective-The accuracy of the oYcial statistic on infant deaths in Taiwan has been questioned. This study aimed to survey infant deaths nationwide, to measure associated vital statistics, and compare them with the oYcial statistics to assess accuracy. Design and participants-A nationwide survey of all gestational outcomes occurring at > 20 weeks' gestation over a three day study period (15-17 May 1989) was conducted to collect data from 23 counties and cities nationwide using a two stage data collection procedure. Main results-The survey derived infant death rate was 9.72 per 1000 live births, which was higher than the reported oYcial statistic of 5.71 per 1000 live births. A more detailed examination of data on infant deaths showed that the estimated neonatal death rate of 6.68 per 1000 live births (95% confidence intervals: 3.33, 11.96 per 1000 live births) was significantly higher than the published oYcial statistic of 1.94 per 1000 live births, while the postneonatal mortality of 3.04 per 1000 live births was comparable to the reported statistic of 3.37 per 1000 live births. Conclusions-Thisstudy empirically documented the underregistration of infant deaths in Taiwan, particularly those occurring during the first 27 days of life. (J Epidemiol Community Health 1998;52:289-292)

No abstract