Let me be clear about the fact that I consider IHC to be one of the greatest developments in Pathology—ever. Indeed, Dr. Albert Coons, the originator of the technique (1), is one of my personal heroes. Nevertheless, the technology is now being used in ways that were never intended by its early proponents. IHC was, and is, a powerful method for determining cellular lineage, and for the diagnostic separation of many biologically-dissimilar lesions that resemble one another morphologically. Having said that, it must also be understood that definite exceptions to such generalities exist. For example, IHC is only marginally useful in distinguishing between sarcomatoid carcinoma and sarcomatoid mesothelioma (2). Even more importantly, with a few selected concessions, IHC is NOT a reliable tool for separating benign from malignant tumors. Moreover, it only crudely (at best) substitutes for formal biochemical or molecular evaluation of cellular products and constituents that affect prognosis (3). Finally, to my way of thinking, it is appalling that many pathologists presently try to substitute the indiscriminate use of IHC for morphological proficiency in the clinical sphere. Rather than mining the histological nuances of sections stained with hematoxylin & eosin, a large and often randomly-chosen panel of immunostains is ordered almost reflexively by such people. This is intended to have an effect like that mentioned in the Bible in Ecclesiastes 11: 1—“Cast your bread on the surface of the waters, for you will find it after many days.” In other words, the casting of many stains on a case should return good things to the person doing the casting. In fact, the practice of randomly procuring immunostains usually produces the opposite result—the bread gets soggy and sinks, and no answer is forthcoming.
As scary as the field of Statistics can be to physicians, it is extremely valuable at the same time. And, principles of Bayesian statistics have a direct bearing on the topic introduced above. The first is the concept of “prior probability.” As defined in Wikipedia (4), the prior probability (“prior”) of an uncertain quantity is the probability distribution that would express one's beliefs about that quantity before other evidence is taken into account. For example, the prior could be the probability distribution attached to the relative likelihoods of selected differential diagnoses. After additional information (e.g., IHC results) is integrated into the analysis, a “posterior probability” (5) emerges.
However, that sequence of events will only be beneficial to refining the diagnosis IF the data used to produce the posterior probability are materially useful. If they are instead random or internally-contradictory, no conclusion can be drawn from them. Therefore, returning to our hypothetical case setting, one could conceivably obtain 40 various immunostains and still not derive useful input from the results. The key element here is obviously the ability to refine the differential diagnosis properly by thorough microscopic examination and clinicopathologic correlation. One can then choose a limited panel of immunostains that is targeted to the specific questions being asked.
What about the converse situation, in which histological assessment is itself dispositive but immunostains are procured anyway, “just to make sure?” Indeed, that practice has now become rife in dermatopathology in the analysis of pigmented skin lesions (6), with the routine but inexplicable application of stains for such markers as melan-A, Sox 10, and HMB45 to virtually all cases regardless of their microscopic appearances. In that setting, let us consider the following scenario—a melanocytic lesion has the classical appearance of melanoma in-situ, with no hint whatsoever of dermal involvement morphologically, but reflexive immunostains appear to show rare melanocytes in the corium. What to do? The posterior probability attached to the case is now less in favor of the correct diagnosis than the prior probability was. Rather than label the tumor as invasive, the wise observer would ignore the problematic IHC information. Indeed, he or she would not have obtained any immunostains at all in the first place. Other examples abound—for instance, cases of histologically-prototypical infiltrating lobular breast carcinoma that are stained reflexively for E-cadherin. In 15-20% of those lesions, immunoreactivity for that marker will be observed unexpectedly (7). Does such a result mean that the “correct” diagnosis is actually that of ductal carcinoma? No, it does not.
So, we come to the denouement of this discussion—the urgent need for experiential judgment and common sense in pathologic diagnosis. A study done by my colleagues in 2012 is germane to that statement. Shah et al. (8) evaluated 200 selected referral cases seen at our institution, which had been subjected to IHC. Two of the three pathologist-authors were experienced people with keen morphological skills. In comparing final diagnoses with those of the submitting pathologists—who comprised a heterogeneous assemblage regarding their practice settings and levels of experience—the level of agreement was 98% between the 2 groups. However, IHC was felt to be necessary by the authors in only 11% of cases, whereas the originating pathologist(s) had obtained immunostains in up to 38%. With potential motivations for the latter finding left aside, one can rightly conclude that the practice-environment and the level of histodiagnostic confidence were major contributors to this difference.
In future, we will increasingly face the need to economize and streamline procedures in anatomic pathology. The CMS has already begun a steady decrease in its reimbursement for IHC, and other technologies will undoubtedly follow suit. The best thing that we can do in response is to train ourselves well in general Medicine and in classical morphology, to optimally use the wealth of information represented in those two disciplines. The degree of “prior probability” that we can provide to the healthcare system is enormous in that context.
REFERENCES
- https://en.wikipedia.org/wiki/Albert_Coons
- Doyle LA: Non-mesenchymal mimics of sarcoma. Surg Pathol Clin 2015; 8: 493-513.
- Wick MR, Swanson PE, Marchevsky AM: Evidence-based practices in applied immunohistochemistry: dilemmas caused by cross-purposes. In: Evidence-Based Pathology & Laboratory Medicine (Marchevsky AM, Wick MR, Eds), Springer, New York, NY, 2011; pp. 261-296.
- https://en.wikipedia.org/wiki/Prior_probability
- https://en.wikipedia.org/wiki/Posterior_probability
- Naert KA, Trotter MJ: Utilization and utility of immunohistochemistry in dermatopathology. Am J Dermatopathol 2013; 35: 74-77.
- Chan JK, Ip YT, Cheuk W: The utility of immunohistochemistry for providing genetic information on tumors. Int J Surg Pathol 2013; 21: 455-475.
- Shah AA, Frierson HF Jr, Cathro HP: Analysis of immunohistochemical stain usage in different pathology practice settings. Am J Clin Pathol 2012; 138: 831-836.