Computer-Assisted Decision Support in Medical Oncology: We Need It Now

Get Permission

In the uptempo pace of cancer care, it would be comforting to have greater confidence that your decisions were not only quick but personalized, evidence-based, patient-centric, and optimal. Computer-assisted clinical decision support in routine oncology care is coming—for many, the sooner the better.

Andrew D. Seidman, MD

Today’s medical oncologist is increasingly challenged to stay current with the latest developments in cancer treatment. I have been fortunate to speak with many oncologists over the past quarter-century on how professional life has evolved since the 1990s. These conversations have left me with a sense of appreciation for how fortunate I am in being able to exist as a one-trick pony—that is, a “master of one” (type of cancer). In contrast, the “jack of all trades” medical oncologist may see 20 or more patients daily with breast, colon, lung, prostate, pancreatic, ovarian, renal cell, and bladder cancers, as well as melanoma, lymphoma, multiple myeloma, and more.

Molecular subclassification of cancers is evolving rapidly. Genomic testing has become more essential to patient management, where “actionable” mutations, deletions, and amplifications connect to agents that improve outcomes. Thus, a significant layer of complexity is added to already multilayered/nodal treatment algorithms (consider the number of footnotes the next time you peruse the National Comprehensive Cancer Network [NCCN] Guidelines®).

Apart from providing cancer care with empathetic, psychosocial support to patients and their families, medical oncologists are increasingly challenged with the transition to and improvement of electronic notes, orders, prescriptions, bills, and toxicity reporting forms. There is a need for time to communicate with your consultants and referring physicians and to teach the fellow, resident, or student rotating through the clinic.

Generalist vs Specialist

Consider the limited amount of time that the more general oncologist may have to contemplate an indicated change in systemic therapy for Mrs. D, a patient with hormone receptor– and HER2 receptor–positive metastatic breast cancer to bone and lung, recurring 13 months after the completion of adjuvant chemotherapy and trastuzumab (Herceptin) while she was on antiestrogen therapy. Indeed, complex decisions are often made quickly in the typical get-it-all-done-in-20-minutes-per-patient clinic, and toward this end, it is useful to have authoritative expert consensus panel–generated guidelines and pathways (eg, from the NCCN, ASCO, St. Gallen) to expedite the decision-making process. Guidelines can provide general broad guidance in decision-making based on disease stage and biologic subtype.

The super-specialist might be able to make a quick and personalized decision that is more nuanced than one informed by guidelines alone. This enhanced level of decision-making is an amalgam of the experience of having a larger volume of cases of the one disease being treated by the physician and the consideration of often not-so-obvious key attributes of the case that go beyond those considered by guidelines. Does the patient have symptomatic or asymptomatic bone metastases? Is she at risk for an impending fracture? Are the liver metastases limited or diffuse? How large? Symptomatic? What was the relapse-free interval from prior adjuvant therapy? Has a repeat biopsy been performed? What is the current status of estrogen receptor and HER2? Does the patient have residual cardiac toxicity or neurotoxicity from prior adjuvant therapy? What logic applies here? What clinical trials exist, and what clinical trial data can one draw upon to inform this decision? CLEOPATRA? EMILIA? PERTAIN? PALOMA-1? BOLERO-2?

Power of Computers

There are many groups that endeavor to leverage the power of computers to enhance the quality of treatment decision-making in medicine, and specifically, in oncology. The assumption, and indeed, the essential motivation for these efforts is not simply that they will enable the harried physician to make better decisions in less time, but that they will actually help our patients live longer and better and maybe even result in cost savings.

In a recent overview evaluating information technology–based decision support systems in medicine, 53 published papers were identified over the past 3 years.1 The most represented area was cancer (12%), yet patient participation in these decision support systems was limited (9/53 studies). The authors concluded that “systems focusing on patient-centered care still do not seem to be available at large.”

When a guideline-based clinical decision support system is employed, it may generate propositions that do not match the recommendations of clinical practice guidelines, leaving the physician with the dilemma of following either the decision support system recommendation or the clinical practice guidelines. In a recent report of oncologists using the OncoDoc2TM decision support systems for the management of breast cancer patients (n = 394 decisions), global decision support system compliance was 75.4%, whereas compliance with the clinical practice guidelines was 86.8%. Errors in patient data entry resulted in nonrecommended propositions in 21.3% of decisions,2 suggesting the need for an automated as opposed to a human user interface–based system. Indeed, the potential for “e-iatrogenesis” is real in the health information technology (HIT) era.3

The ability of many stakeholders to collaborate to leverage health information technology to drive discovery and improve patient care is characteristic of efforts such as ASCO’s CancerLinQTM, a rapid learning system for cancer care.4,5 Given that a condition such as metastatic breast cancer behaves like a story with many chapters, optimizing each therapeutic choice along the way should lead to improved outcomes and cost savings.6

The Watson Project

IBM Watson is a cognitive computing system that ingests data from many sources (the “corpus”) and is not limited by volume or memory. It requires training by a team of human experts to learn how to interpret the information and is being developed in oncology by several teams,7 including one at Memorial Sloan Kettering (MSK). After cycles of deidentified actual patient data as well as manufactured cases are run through Watson, output is carefully reviewed and analyzed by teams of MSK experts (one-trick ponies, to be sure) and IBM information technology experts working closely together.

The cognitive abilities of IBM Watson allow it to adapt to unique patient scenarios, interact with the user, and provide context for decisions, among other capabilities. Soon, Watson will rapidly ingest the key attributes of a patient’s case by employing natural language processing to read a consultation note and other components of the electronic medical record, a complex linguistic/informatic endeavor that is currently being refined. It will assist the busy general oncologist to make a more nuanced treatment decision based not only on the corpus, but upon unique curated key attributes of each patient’s case not captured by current consensus guidelines. It will also factor in patient preference as a key attribute, particularly when considering palliative therapies.

These decisions are more likely to align with the refined decisions that might be made by the one-trick pony oncologist. Where guidelines can generate broad treatment options, Watson output can be more granular and patient-specific.

Case Continuation

Returning to Mrs. D, whose cancer has recurred 13 months after she completed adjuvant anthracycline, taxane, and trastuzumab-based chemotherapy while she was on tamoxifen, you consider the following factors:

(1) The patient is 51 years old and amenorrheic since chemotherapy. Her serum estradiol level is in the postmenopausal range, as are her pituitary gonadotropins. She has two pulmonary nodules measuring < 2 cm each. She has four osteolytic bone metastases—two in the ribs, one in the L3 vertebrae, and another in the ilium, remote from her acetabulum. She has no respiratory symptoms; she has mild pain relieved with acetaminophen. She has residual grade 1 peripheral neuropathy from prior taxane therapy. Laboratory studies reveal mild renal insufficiency. A computed tomography (CT)-guided lung biopsy demonstrates metastatic breast cancer that is 20% estrogen receptor–positive, progesterone receptor–negative, and HER2 3+ by immunohistochemistry (her primary breast cancer was 70% estrogen receptor–positive, 30% progesterone receptor–positive, HER2 3+). She is not interested in deeper molecular profiling of her tumor at this time or participation in clinical trials.

(2) Data from relevant clinical trials would inform her care in an evidence-based manner.

(3) The patient states that her daughter’s wedding is in 3 months, that she wants to feel as well as possible, and that she has a strong aversion to losing her hair again, now that it has grown back following adjuvant chemotherapy.

(4) It is 2:00 PM, you have seen 16 patients already today, and you are running an hour behind schedule. Your nurse asks you to speak with the radiation oncologist who has been holding on the phone regarding another patient’s stereotactic radiation for brain metastases set to start today. Your office assistant informs you that Ms. T, the patient you saw an hour earlier, is now experiencing a hypersensitivity reaction in the chemotherapy suite. Your daughter just texted you again [“OMG”] to remind you that her tuition bill for fall semester has not yet been paid, and “Daddy can you log on and take care of this now?

You are thinking about the low estrogen receptor expression and that despite Mrs. D’s mild symptoms, she does have visceral disease. You are pretty sure that she is in menopause, although she reported some spotting on tamoxifen recently. Maybe it’s not the time to pursue antiestrogen therapy.

You know that your office manager will need to get insurance approval for either pertuzumab (Perjeta) or ado-trastuzumab emtansine (Kadcyla). You are thinking about prescribing denosumab (Xgeva), not zoledronic acid, given the mild renal insufficiency. There is no need for orthopedic consultation, as the lytic iliac lesion is not close to the acetabulum. Perhaps vinorelbine with trastuzumab and pertuzumab will be optimal, so that she will not need to wear a wig at her daughter’s wedding?

As you return from the chemotherapy suite (where Ms. T’s allergic reaction has fortunately resolved, and you make a mental note to switch her treatment to albumin-bound paclitaxel in the next cycle), you pick up the phone to speak with the understandably impatient radiation oncologist. You simultaneously notice your daughter’s second e-mail with the red exclamation mark and see that your lunch is sitting neglected by your workstation—and you wonder if you had indeed ordered a wilted kale salad, as your soup gets cold.

You reenter the examination room to discuss the possible treatment options. You patiently absorb Mrs. D’s concerns and fears and spend time explaining treatments that come to mind. Thirty minutes later, you have mutually decided on a plan; you have entered the orders, scheduled follow-up appointments, submitted prescriptions, and are now 90 minutes behind schedule to see your 17th patient of the day.

Coming Soon

Twenty years ago, there were 68 controlled trials of clinical decision support systems, with 19 of 26 studies on medical care demonstrating a benefit in physician performance and 6 of 14 studies showing a benefit in patient outcomes. Computational power has expanded enormously since this time. It is not unreasonable to expect that by 2020, medical oncologists will be using clinical decision support systems in real time in the clinic. Treating cancer is proving to be more complicated than winning at Jeopardy!8 but significantly more gratifying.

Forty-five years ago, Dr. William B. Schwartz opined that computers gave little hope “of dealing with the difficult challenge of maintaining a high level of physician competence in the face of a continued expansion of medical knowledge that tends to widen progressively the gap between what a doctor should know and what he [or she]can retain and utilize.”9 In the uptempo pace of cancer care, it would be comforting to have greater confidence that your decisions were not only quick but personalized, evidence-based, patient-centric, and optimal. Computer-assisted clinical decision support in routine oncology care is coming—for many, the sooner the better.

You finally get to your kale salad and adroitly avoid the late fees on your daughter’s tuition. Go home. ■

Disclosure: Dr. Seidman reported no potential conflicts of interest.


1. Quaglini S, Sacchi L, Lanzola G, et al: Personalization and patient involvement in decision support systems: Current trends. Yearb Med Inform 10:106-118, 2015.

2. Bouaud J, Spano J-P, Lefranc JP, et al: Physician’s attitudes toward the advice of guideline-based decision support system: A case study with OncoDoc2 in the management of breast cancer patients. Stud Health Technol Inform 216:264-269, 2015.

3. Weiner JP, Kfuri T, Chan K, et al: “e-Iatrogenesis”: The most critical unintended consequence of CPOE and other HIT. J Am Med Inform Assoc 14:387-388, 2007.

4. Abernathy AP, Etherege LM, Ganz PA, et al: Rapid learning system for cancer care. J Clin Oncol 28:4268-4274, 2010.

5. Sledge GW, Hudis CA, Swain SM, et al: ASCO’s approach to a learning health care system in oncology. J Oncol Pract 9:145-148, 2013.

6. Luke J, MIrkin J, Bach P: Improving quality and addressing the rising costs of cancer care: Two birds, one stone. J Oncol Pract 7:402-404, 2011.

7. Kantarjian H, Yu PP: Artificial intelligence, big data, and cancer. JAMA Oncology 1:573-574, 2015.

8. Markoff J: Computer wins on ‘Jeopardy!’: Trivial, It’s Not. The New York Times, February 16, 2011.

9. Schwartz WB: Medicine and the computer: The promise and problems of change. N Engl J Med 283:1257-1264, 1970.