Role of Evidence Based Medicine in Diagnostic Surgical Pathology

Evidence-based medicine (EBM) has been defined as “the conscientious, explicit, and judicious use of current best evidence in making decisions about the care of individual patients” or as “the integration of best research evidence with clinical expertise and patient values”. It is an evolving discipline that applies analytical and quantitative methods to evaluate the validity of available medical information, with the overall goal of identifying scientifically-sound data or “best evidence”. This evidence is integrated to improve medical practice through clinical guidelines and other tools that are used for education, standardization of care, quality initiatives and coverage decisions. The ideas of EBM have spread rapidly through medicine during the past decade and are recently eliciting a growing interest in Anatomic Pathology and Laboratory Medicine.

Basic Concepts of Evidence-Based Medicine

EBM investigators attempt to identify the best current and relevant research information available for a particular problem and to integrate the “evidence” into guidelines, rules or other tools that will assist medical practitioners in their daily practice.
Basic Process for the identification of best evidence and its integration into guidelines, rules or other protocols.

1. Formulate specific questions regarding the diagnosis, prognosis, causation and/or treatment of individual patients with a particular clinical problem
2. Search for specific information in the scientific literature
3. Appraise the internal and external validity of the available evidence, and its impact, applicability and usefulness in daily practice
4. Incorporate “best evidence” from several reliable sources along with personal clinical exexperience into” guidelines, rules or other protocols
5. Evaluate the effectiveness and efficiency of those “Evidence-based” recommendations

Bayesian approach to the analysis of data : influence of the prior probability of findings of interest.

Descriptive statistical tests offer limited information about other features that can influence the outcome of observational studies, such as the prevalence of a disease within the population study and in the control group and the prior probability of a finding. For example, it is well known that lymph node status has a statistically significant prognostic significance in most patients with cancer. However, in patients with Stage IV neoplasms who have a high “prior probability” of dying from their disease, the prognostic value of the feature lymph node status is probably rather limited. These considerations are intuitively used in daily practice by most pathologists, but there are few, if any, available evidence-based guidelines or other protocols that take into consideration the prevalence and prior probability of various findings into the selection and/or interpretation of diagnostic features, immunostains or other ancillary tests in Surgical Pathology. Another consideration that has not been addressed in most observational studies in Pathology is the need to divide the data into “training” or “testing” sets (“study” and “holdout” cases) in research projects attempting to derive classification or prognostic models. Most clinico-pathological categorization has been based on data derived from analyzing 100% of the data from study groups and control groups with descriptive univariate, and less often multivariate statistical methods. However, multiple studies using Bayesian methods have shown that models derived by the use of 100% of a dataset usually have limited external validity as there is a certain element of “circular reasoning” in the modeling methodology.

Evaluating the Quality of Published Studies in the Medical Literature.
Ebell has proposed a system for classifying published medical evidence into 4 levels, with “grade I” being the best (most reliable). Grade I studies are those that include data validated with a “test” group that is from a different and distinct population from the “training” cohort. Grade II studies report data that are obtained from the same population, the members of which are divided into independent “training” and “validation” subsets and evaluated prospectively. Grade III analysis also include “training” and “validation” subsets from the same population, but data are collected contemporaneously rather than prospectively. Grade IV studies are those in which the “training” group is also used as the “validation group”. According to this scheme, most studies in the pathology literature would probably be classified as Grade IV, the most vulnerable to external validity problems.

Integration of “best evidence” from the literature with personal clinical experience into “evidence-based” guidelines, rules or other protocols.

Advocates of EBM have attempted to organize “best evidence” from the scientific literature and their own experience into algorithms, protocols, guidelines or “rules” that guide individual patient care by practitioners. Pathologists may benefit from emulating this approach, in future efforts at constructing “patient-based” prognostic and predictive models. For example, immunostains are most often used to distinguish between various neoplasms in a descriptive manner. Studies using immunostains in the pathology literature usually list the percentage of lesions that label for particular epitopes, as well as the sensitivity, specificity and predictive values of such markers in narrow morphological contexts. However, few studies have assessed these data with meta-analysis or calculated likelihood ratios (LR) or other probabilistic measures as applied to panels of markers in selected differential diagnoses. At an even more basic level, the relative statistical values attending particular morphological findings has seldom been analyzed in the same fashion. In contrast, several prognostic scoring models or “rules” that integrate multivariate pathological, clinical, imaging and other information are being developed by other specialists. For example, Kattan and associates have developed pretreatment nomograms that combine clinical and pathological data from prostate cancer patients and predict 5-year probability of metastasis.

Data collected from surgical pathology specimens can also be integrated with the use of “tools for reasoning with uncertainty” such as rule-based expert systems, multivariate statistics, Bayesian belief networks and neural networks.