The Art of Algorithms
How can an algorithmic approach to diagnosis strengthen the practice of pathology?
Pranav Pramod Patwardhan | | Longer Read
At a Glance
- An algorithmic approach to diagnosis promotes a logical, sequential, and organized thought process
- Diagnostic algorithms can be handy resources and allow pathologists to share their knowledge easily
- They are especially useful for trainees, who can not only make use of existing algorithms, but also learn to create their own
- Such algorithms move from history and clinical information all the way to immunohistochemistry and molecular diagnostics
Every pathologist knows that there is an element of art to the science of diagnostics. Not every disease presentation is typical. Sometimes, an important finding is barely captured on the edge of a slide, or buried in a patient’s clinical history. Other times, an abnormality in a biopsy may lead you to not only a primary diagnosis, but also an unrelated second observation. But with so much information to corral, how can you progress logically through the steps from sample to diagnosis, and how can you ensure that you are taking note of every important finding along the way?
I recommend that every pathologist and trainee develop an algorithmic approach to the slides and samples we see every day. And although I create algorithms for my own use in the clinic, my main motive in sharing them – and in writing this piece – is to stimulate my colleagues and trainees in the discipline and help them develop an analytical approach to pathology.
But what exactly are these diagnostic algorithms? And what features make them complete, meaningful, and useful in training and in our daily practice.
Foundations of an algorithm
To develop a diagnostic algorithm for any lesion or tumor, it’s important to establish your first steps. Just as the evaluation of any slide should begin with a few simple questions, your approach to any case begins with the gross examination and the clinical details you receive from the operating surgeon. Different key points in the history and clinical exam may be important for different organ systems, so relevant positive and negative findings are commonly placed at the first level of any algorithm I make.
For example, in the evaluation of a central nervous system tumor, the patient’s age and the site of the lesion should be the first things you ask about; they can give you specific hints as to what you might expect to find in the microscopic evaluation. A lateral ventricular tumor should make you think not only of the common ventricular tumors, but also of subependymomas, subependymal giant cell astrocytomas (SEGAs), and central neurocytomas. If you also know that the patient has a history of tuberous sclerosis, you should start your microscopic evaluation with a strong probability of SEGA in mind. Meningeal and posterior fossa-based tumors are also cases in which the site may lead you to a diagnosis even before you see the slide under the microscope.
When dealing with a bone lesion or tumor, the age of the patient, the site of the lesion, and the radiological information should be the first step in your algorithmic approach. If you’re working on an algorithm for hematolymphoid malignancies, you should consider physical examination findings like hepatosplenomegaly, lymphadenopathy, back pain (in a suspected plasma cell dyscrasia), and peripheral blood smear evaluation. By now, you may have noticed a pattern: clinical details and supportive radiological information are the two key factors that should inform the initial steps of an algorithm for any pathological lesion.
Once that information is established, gross examination is the logical next step. New residents are often fascinated by the advent of immunohistochemistry (IHC), molecular pathology, and other novel approaches – but may sometimes miss minute details on the gross. Systematic gross examination in organs like the adrenal glands can be very rewarding. There, bilateral multifocal lesions will make you think of metastasis; a single, well-circumscribed nodular lesion will lead you toward a diagnosis of adenoma; a dusky brown color of the tumor will point to pheochromocytoma; and anything large, hemorrhagic, or necrotic will make you think of adrenocortical carcinoma. Relatively well-circumscribed cystic lesions with only a few hemorrhagic areas mixed with yellowish-white tissue will hint at the possibility of myelolipoma or angiomyolipoma.
The “2C”s and“2S”s of gross examination – color, consistency, size, and shape – are the key points to take away from the exam. Gross clues are important in diagnostic algorithms for nearly all organ systems, but they hold particular importance in genitourinary, ovarian, and gastrointestinal pathology. Fine needle aspiration findings should also be included in the algorithms for hematolymphoid malignancies and tumors of the thyroid and salivary gland.
No stain, no gain
Microscopic examination still benefits from classic H&E staining, which continues to provide diagnostic insights. As you’re examining a tissue sample, you must keep in mind new diagnostic entities with greater prognostic significance or treatment relevance; differentiating between two tumors formerly considered a single entity could have a vast impact on the patient’s treatment or expectations.
This step in the algorithm is not to be taken lightly. Pathologists should consider each cell component – nucleoli, nuclear membrane, chromatin, cell inclusions (if any apart from the surrounding vasculature), inflammatory infiltrates, and mitosis – when narrowing down a differential and a final diagnosis. A good example is when differentiating oncocytomas from chromophobe renal cell carcinomas on the basis of nuclear membrane irregularity and perinuclear halo.
Special stains and their interpretations should be included in the algorithms whenever relevant; for example, in liver biopsy interpretation, bone marrow examination, or suspected pituitary adenoma. It’s also good practice at this stage to add histological details on H&E that may be associated with a particular genotype or mutation. A few examples:
- Tumor cells in hereditary leiomyomatosis and renal cell cancer show eosinophilic nucleoli with a peripheral halo, arranged in a type 2 papillary pattern.
- Colorectal carcinomas in cases of Lynch Syndrome are more likely to show tumor-infiltrating lymphocytes and less likely to show dirty necrosis, characteristically described in intestinal adenocarcinomas.
- Amyloid or calcification in a pituitary adenoma would suggest a prolactinoma, paranuclear fibrous bodies a somatotropic adenoma, and Crooke’s hyaline change in the surrounding normal pituitary a corticotroph adenoma.
These histological clues can point you toward particular genetic or functional subtypes of tumors and help correlate your histological findings with the molecular diagnosis later on. In situations where molecular testing may not be feasible, they enable you to provide at least some guidance to surgeons and oncologists.
It’s at this point – not before – that pathologists should begin to consider immunohistochemical and molecular diagnostic details. Although the use of these assays is inevitable in many scenarios, it can be difficult – especially for trainees and new pathologists – to fully understand their relevance without the previous steps. This, then, is the key to the algorithmic approach: it must first lead you to a narrow differential diagnosis on the basis of the history, gross, and H&E microscopic evaluation, and then guide you to the relevant IHC and molecular studies for a conclusive diagnosis wherever possible.
A good IHC evaluation should ask four questions: i) Which cells take up the marker?, ii) Which component of the cell shows positivity?, iii) What is the pattern of that positivity?, and, most importantly, iv) Are your markers working on your internal controls? All of these points are vital to the final step of your diagnostic algorithm.
Additionally, if the IHC or molecular characterization of your tumor is complex, I recommend a separate algorithm or “sub-algorithm” for this step. The diagnostic evaluation of lymphomas, for example, is somewhat complex. It helps pathologists, especially those new to the discipline, understand the significance of each marker in turn. Let’s say you have a case of B cell lymphoproliferative disorder in which H&E evaluation suggests a diagnosis of chronic lymphocytic leukemia (CLL) or small lymphocytic lymphoma (SLL). Your algorithm would recommend first evaluating CD5 and then CD23. Positivity of both markers indicates CLL, whereas a negative CD23 result suggests mantle cell lymphoma. If both markers are negative, the algorithm should recommend testing others (such as CD10, CD25, or CD103). In this example, the diagnostic process is a relatively common diagnostic process – but you can create and use this kind of sub-algorithm in any similarly complicated case.
When conducting these analysis, always bear in mind that the stroma – the “climate” of the tissue on the slide – can sometimes lead to an unrelated diagnosis if you observe it closely. For instance, in a duodenal biopsy, the absence of any plasma cells should make you think of common variable immunodeficiency (CVID); diffuse generalized hyaline arteriolosclerosis should alert you to the possibility that the patient might be hypertensive; ulcerative or exudative lesions with “volcano”-like inflammatory infiltrates should make you think of pseudomembranous colitis! Nowadays, whenever I make a checklist for duodenal biopsies, I invariably include a point on CVID after I finish describing the luminal features (including parasites like Giardia) and mucosal changes (which may point to issues, such as celiac disease).
An educational tool
These “personal diagnostic algorithms” are useful for any pathologist at any career stage, but I find them particularly valuable for residents and trainees. Why? Because working in this way helps them to develop an analytical approach toward any case evaluation. It fosters a thought process that uses logical, sequential reasoning to arrive at a final diagnosis – exactly what we intend to achieve when we train new students. The eye cannot see what the mind doesn’t know!
Though the details at times appear complex, I encourage all of my colleagues to develop a sharp eye for the “catchy” diagnostic points of any lesion they encounter, whether in real life or in the literature. Always think about how you would differentiate those entities from other similar ones. And, if the knowledge you pick up is not already present in the literature, it may be a valuable resource for dissemination amongst your peers.
The diagnostic algorithms we build up using both published data and our own experience are extremely useful tools for us, and for those we train to follow us. I share my algorithms with my trainees, but I also urge them to create their own, which not only provides them with the tools themselves, but also with the thought processes that arise from pursuing a logical approach to diagnosis. Remember that all of histopathology is based upon a simple, algorithmic principle:
- Is the tissue normal or abnormal?
- If it is abnormal, is it neoplastic or non-neoplastic?
- If it is neoplastic, is it benign or malignant?
- If it is malignant, is it primary or secondary?
And, of course, don’t forget the value of a good report after you’ve put in the work to reach a diagnosis. Frame it carefully, word it well, and make sure that the final report is just as logical as your diagnostic process!