Amir Momeni

Imputation and Missing Data

The presence of missing data is a big challenge for statisticians, especially if the distribution of the missing values is not completely random. Analysis performed on datasets with missing data can lead to erroneous conclusions and significant bias in the results.

Validation of New Tests

Validating new tests is a recurring problem that pathologists face. The field of diagnostic medicine has a high rate of innovation and the laboratories need to adapt and implement new tests to address clinical needs and remain relevant. Validation ensures that the test performs as expected and satisfies the requirements of the lab, manufacturer, and regulatory bodies.The validation process requires the test to fulfill the requirements of its specific intended use through objective measures. The tests need to satisfy these requirements and they need to do it consistently. This process has three steps: define performance goals, assess error and finally compare the error with the goals. In this chapter, we will explain the steps involved in validation of new tests and explain the objective measurements that are needed for validation.

Probability and Probability Distribution

The theory of probability and statistics underlies all quantitative assessments relevant to pathology. Probability is a concept that occurs due to randomness. A random event or experiment has two main components of interest: the first is the “outcome” which is the result of the event that is being recorded. The second is “parameter” which is a constant in the experiment which can affect the outcome. Most events in biology are random and thus are governed by laws of probability, and, consequently, laboratory medicine where our concern is to measure these biological events is also governed by randomness and probability. Pathology and laboratory medicine employ the concepts of probability to provide us with a posttest probability of someone having a disease based on their test result, pretest probability, and the likelihood associated with the test. Another important concept is probability distribution which is the probability of occurrence of different possible outcomes of a random trial or experiment. Test characteristics such as “confidence interval,” “mean,” “reference range,” “error,” etc. are all determined using probability distribution.

Why Every Pathologist Needs to Know Statistics

Statistics is integrated in many aspects of practice of pathology; we use statistics in design, implementation, and interpretation of diagnostic and prognostic tests. Tests are only meaningful because they can distinguish diseased status from the unaffected status. This distinction can only be made with acceptable certainty if statistical process has been followed in the design and interpretation of the test; characteristics such as sensitivity, specificity, and accuracy are derived from statistical methods. The application of statistics in pathology encompasses clinical, technical, and administrative aspects of practice of pathology.

Designing Diagnostic Studies

Research and investigation are important components of the practice of pathology. Pathologists are mainly involved in two types of clinical research: diagnostic or prognostic. Diagnostic research is conducted to improve diagnostic procedures and tests with the aim of improving diagnostic accuracy. Prognostic research mainly aims to identify and quantify factors that dictate the prognosis in patients.

Statistical Concepts in Laboratory Quality Control

Quality control ensures that the performance of a test is within the limits set by validation experiments as well as the requirements of the lab and regulatory bodies. The goal of quality control is to minimize variability and maximize accuracy and precision; this requires measurements of quality metrics and interpretation and analysis of these quality metrics by statistical methods.

Assessing Diagnostic Tests

Pathologists need to assess diagnostic tests and to decide whether to incorporate a test into their practice and decision-making process. This requires them to understand and interpret the technical and clinical characteristics of the tests. The technical utility of tests is dependent on their technical accuracy and precision which are determined by identifying and quantifying measurement error. Clinical utility of diagnostic tests is dependent on the ability of the test to correctly distinguish diseased state from the unaffected state; the clinical accuracy is defined using terms such as sensitivity and specificity.

Cross Tabulation and Categorical Data Analysis

Often, we have questions about associations of events or variables with each other or their correlation with each other. For example, in pathology we commonly face the question of association of a test result with a disease status. In statistics, the process of testing the association between events is called hypothesis testing. If the variables are categorical (i.e., they can only assume finite discrete values), a common approach to hypothesis testing is to employ cross tabulation.

Critical Appraisal of Diagnostic Studies

It has been shown that diagnosis utilizes approximately 5% of healthcare costs, yet 60% of the clinical decision-making process is dependent on the diagnosis. Since approximately 40,000–80,000 hospital deaths per year in the United States are attributed to misdiagnosis, reduction of misdiagnosis is now being recognized as a major goal in patient safety efforts. Many of these deaths are preventable deaths that can be avoided if a correct and timely diagnosis is made. Pathology and laboratory medicine in their major roles as a source of diagnosis or a major contributor to the diagnostic process require a push toward more accurate and precise testing and diagnosis, and this requires integrating the best available evidence into the everyday practice of pathology.

Comparing Sample Means

In pathology and laboratory medicine, there are many tests that provide quantitative and continuous results. In fact, many qualitative, categorical test results are derived from quantitative results that are dichotomized for ease of interpretation. Continuous variables can assume any value in a real number interval. Statistical inference of continuous variables is very important, for example, comparing the quantitative results of a test between two groups of patients requires statistical tools that can deal with continuous variables.