Development and Implementation of De Novo Clinical Decision Support Rules at a Tertiary Cancer Center


Aymen Elfiky: Dana-Farber Cancer Institute, Boston, MA; Brigham and Women’s Hospital, Boston, MA; Harvard Medical School, Boston, MA.

Marina Kaymakcalan:  Dana-Farber Cancer Institute, Boston, MA.


Aymen Elfiky; Dana Farber Cancer Institute; 450 Brookline Avenue; Boston, MA 02215;


The authors report no financial interests.

Key Words

Abstract: The rising costs associated with cancer care delivery have prompted the health care system to prioritize the principle of value-based care to accentuate the concepts of quality and outcomes. Driven in large part by the observed variability in care delivery and the move toward increasing accountability, clinical pathways (CPs) programs are becoming part of the practice infrastructure for large oncology practice networks, commercial, and clinic-based groups. The authors describe a CP initiative to address the growing variations in care practices with the advent of new treatment approvals for advanced prostate cancer. Through an iterative CP-development platform, a multi-disciplinary group of providers, pharmacists, and quality improvement specialists employed a rapid health technology assessment (HTA) approach to extrapolate and implement a core set of decision support rules for the use of granulocyte colony stimulating factor (GCSF), a common growth factor support medication, as primary prophylaxis for prevention of treatment complications in advanced prostate cancer patients receiving palliative chemotherapy. Using an approach of transparency of rapid HTA methods to facilitate clinical practice consensus, a final set of defined rules for GCSF use was developed for application within a major academic medical center and across its affiliated community satellite sites. In addition to demonstrating the role of heuristics, provider intuition, and behavioral economics, fundamental organizational limitations and data challenges to sustaining and scaling actionable informatics-based initiatives are highlighted. 

Key words: knowledge creation, information management, clinical pathways, quality improvement, evidence-based medicine, organizational processes

Citation: Journal of Clinical Pathways. 2016;2(3):39-45. 

Received March 14, 2016; accepted March 28, 2016.


Related Content

Effect of Route of Bortezomib Administration on Treatment Outcomes in Previously Untreated Patients with Multiple Myeloma: A Retrospective Analysis from US Community Oncology Practices

Integrating Electronic Health Records into Clinical Pathways for Radiation Oncology Reduces Waiting Times


With rising costs of cancer care, payment and practice reforms are commanding due attention and pressure for action. Traditional cost drivers include chemotherapy, emergency room visits, hospital admissions, and aggressive end-of-life care. Increasingly, these are being compounded by the costs of modern therapies and the drive toward precision oncology.

With cancer being a leading cause of death in the United States,1 the rising costs associated with cancer care delivery have prompted the health care system to prioritize the principle of value-based care2 to accentuate the concepts of quality and outcomes. The coordination of information processing, communication, and management remains a cornerstone of health care delivery. Yet, the health care system’s effort to further the integration of information technology (IT) and communication continues to reveal long-standing gaps and challenges to its investment.3 Paradoxically, the push for a health care IT (HIT) infrastructure has impeded some aspects of “value” attainment and, as a result, HIT has not yet realized its potential for cost benefits. Moreover, the priorities of enabling timely provider reimbursements and demonstrating basic compliance with the Health Information Technology for Economic and Clinical Health (HITECH) Act’s federally mandated “meaningful use” initiative have not included incentives for innovation or tailored care delivery.4

Advances in HIT are important from a health technology assessment (HTA) perspective as well. Given the pace of scientific insights, new therapeutics, and evolving treatment paradigms that are characteristic of oncology, there are both the necessity and the opportunity to leverage existing HIT and specific data assets to overcome the limited availability of prospective, evidence-based data. Increasingly, in circumstances where selected impacts are of particular interest, the demand for HTA by health care decision-makers has involved requests for faster responses to help inform emergent decisions.5 Among the reported applications for rapid HTAs on the order of weeks are coverage decisions, capital funding, formulary decisions, treatment referrals, guideline formulation, and indications for further research.5

Treatment of advanced cancer patients for whom goals of care are by definition palliative, as opposed to curative, has been shown to be characterized by unexplained, significant regional variation in the selection of chemotherapy as well as ancillary support medications and processes.6-8 These variations are regional as well as associated with age, race, and socioeconomic factors.6-8 The result of such variations is wide differences in costs but without a corresponding difference in the quality of care.6-8 Specifically, the absence of focused evaluations and standardized use of expensive compounds such as hematopoietic growth factors or bone remodeling agents can account for significant variations between providers and patients within the same institution. This was the case at the Lank Center for Genitourinary Oncology (GUC) at the Dana-Farber Cancer Institute (DFCI), where the series of new treatment approvals for advanced prostate cancer beginning in 2010 through 2012 prompted the need to address the growing variations in care practices across the center. Moreover, the pace of costly drug approvals combined with increasing patient volumes and expansion of clinical operations into the community networks prompted increased attention from payers. As a result, payers have been attributing increasing financial responsibility to patients in the form of copayments, coinsurance, and higher deductibles.9

Driven in large part by this observed variability in care delivery and the move toward increasing accountability, clinical pathways (CPs) programs are becoming part of the practice infrastructure for large oncology practice networks and commercial and clinic-based practice groups.10-14 In providing a framework for reducing variability and facilitating best evidence-based decisions at the point-of-care, CPs represent an ideal platform for translation of HTA outputs to the clinical setting. An iterative HTA process functioning to optimize the CPs in real time is consistent with the principles of total quality management and continuous clinical quality improvement across the care continuum.

Through an iterative CP-development platform within our academic cancer center, a multidisciplinary group of providers, pharmacists, and quality improvement specialists employed a rapid HTA approach to extrapolate and implement a core set of decision support rules for the use of granulocyte colony stimulating factor (GCSF), a common growth factor support medication, as primary prophylaxis for prevention of treatment complications in advanced prostate cancer patients receiving palliative chemotherapy. Using an approach of transparency of rapid HTA methods to facilitate clinical practice consensus, a final set of defined rules for GCSF use was developed for application within a major academic medical center and across its affiliated community satellite sites.


Concordant with fee-for-performance based negotiations between our institution and a private payer organization, pharmacoeconomic assessments were undertaken that highlighted the use of GCSF as primary prophylaxis against risk of neutropenic infections. When presented with the problem of defining criteria for GCSF administration, the focus of the analysis was on the use of GCSF as primary prophylaxis in patients with advanced disease for whom chemotherapy was being given for palliative intent as opposed to curative intent. It was reasoned that primary prophylaxis would be harder to justify from a cost-effectiveness perspective in this setting than among patients treated with curative intent.

The utilization rate of GCSF was highest in the GUC compared with other disease centers at DFCI, with further investigation showing the most variability among advanced prostate cancer patients receiving primary GCSF in conjunction with cabazitaxel (Jevtana®) chemotherapy. As a consequence of this practice landscape, the concern for variably over- or under-treating patients with GCSF, and the clinical quality implications thereof, were brought to light.

We hypothesized that curation of data assets specific to the drug (cabazitaxel), patient population (prostate cancer), and defined clinical trends would highlight a pattern of association from which a set of actionable use parameters could be defined (Figure 1). This hypothesis drove our two-phase rapid HTA approach.

The previous establishment of the Genitourinary-Management and Assessment Pathways (GMAP) initiative by the authors at DFCI provided the necessary infrastructure for the rapid HTA. Beyond a simple tumor-board format, the intended utility of the GMAP was to facilitate an iterative, data-driven, multidisciplinary consensus of the urologists, medical oncologists, radiation oncologists, nursing providers, and pharmacists within the disease center. The result of this consensus was clear, concise pathways that detailed coordinated care activities across the treatment timeline for each clinical state of the disease.

Under the auspices of the GMAP, the up-front management of expectations, approach, and goals of our problem-oriented rapid HTA was outlined to the group and championed by the GUC Clinical Director and the DFCI Chief Quality Officer. The initiative was approved by the DFCI Institutional Review Board as expedited research prior to data collection. 


The trade-off that must be considered with rapid HTAs is that information, although provided in time to guide action, is limited and less certain, whereas comprehensive and more certain information may not be obtained until after the need to make an effective decision has passed.15 Within the context of a rapidly evolving cancer care landscape as well as the precision oncology mandate, the information and data assets that can best be leveraged for rapid HTA are those that are enriched in such a way as to allow for consistent patient segmentation and extraction of actionable clinical differences from the heterogeneity of treatment effects. Examples of this enrichment are cancer type, treatment sequences, dosing, and responses by age, gender, and comorbidities. The rapid HTA approach can be optimized by limiting the scope to fewer types of impact or evidence questions, focusing searches on defined databases, relying on select types of studies (eg, only systematic reviews or only randomized controlled trials), use of shorter and more qualitative syntheses with categorization of results without meta-analyses, and more limited or conditional interpretation of findings or recommendations.6

Our analysis primarily excluded the use of GCSF with agents that had a relatively low risk of febrile neutropenia (<10%), because NCCN/ASCO national guidelines do not support the use of GCSF in such cases.16 Cabazitaxel is a second-line chemotherapeutic agent given to patients that are resistant to first-line docetaxel (Taxotere®) chemotherapy. According to NCCN guidelines, cabazitaxel is considered an intermediate risk regimen for febrile neutropenia (10-20%). Given the lack of existing data or guidelines which speak directly to this clinical scenario, the GMAP platform facilitated extraction and integration of data assets (Figure 2) to delineate prediction features.

In Phase 1, the GMAP’s designated CP architect proceeded to aggregate the various sources of documented data using a relational database that was set up for this purpose. As a precaution against incorporation biases that could ultimately impact the decision rule, an objective a priori candidate list of items/features was used to guide data collection (Table 1). Data sources specific to the treatment agent (ie, cabazitaxel chemotherapy) included published Phase 2 and 3 trial data, published abstracts presented at national conferences, and unpublished clinical and toxicology data and were proactively solicited from the pharmaceutical industry manufacturer. This data aggregation was undertaken with the task of identifying and comparing neutropenic complication rates between patients who did and did not receive primary GCSF growth factor support. Correlative reports of patient specific factors with rates of toxicity were also solicited from the pharmaceutical drug manufacturer.

Data sources specific to the patient cohort (ie, stage IV prostate cancer patients) included the structured data contained within the electronic health record, specific to the GUC practice setting and with strictly defined baseline inclusion/exclusion criteria of patients (Table 2). Specifically, 125 consecutively sampled patients were chosen to avoid introducing selection biases for which adjustments could not be made at later times.

Multivariate regression analyses were performed using the EHR data of the patient cohort to determine association of factors with complication risks.  These derived factors were compared with published data.


The subsequent derivation of valid and actionable clinical decision support rules is fundamentally dependent on the prevention against conscious and subconscious biases as well as data shortcomings.17 Using the objective, contemporaneously identified predictor variables that could be consistently extracted from the medical record, multivariable logistic regression analyses were applied in order to algorithmically derive relevant risk factor (RF) features as a function of the following prompts: (1) How many RF are needed to justify primary GCSF prophylaxis? (2) Are certain RF to be “weighted” more than others? (3) What permutations of RF are associated with highest risk of neutropenic complications?

Notably, the above stepwise approach was preferred over the direct classification of patients as being at high risk of neutropenic complications, given the relatively small overall sample and given the intention to mirror a provider’s heuristics. Similarly, the small sample size prevented binary recursive partitioning analyses given the concern of over-fitting the data. While outside the specific scope of this paper, comprehensive performance (sensitivity and specificity) analyses, including derived receiver operating characteristic (ROC) curves, were summarized and compiled for presentation to the GMAP group. For each of the RF prompts above, in order to maintain transparency with the GMAP group, the CP architect did not set any “lower limit” performance threshold for presentation to the group.  


Phase 2 of the GMAP effort was the orchestrated negotiation of the above results into a final set of consensus-based parameters. Convening the multidisciplinary GUC stakeholders, a modified Delphi method18 was used over a series of two in-person meetings and a teleconference to achieve a consensus. Given the need to tailor rapid HTAs to the particular needs and time constraints of decision-makers, having committed to an iterative process or refinement of CPs over time, we determined that the transparency of rapid HTA methods used to facilitate clinical practice consensus is more important than achieving consensus on any standard HTA approach itself.19 Importantly, the inherent but relative limitations of the data analyses above were preemptively stated to the group. At the same time, however, the fundamental purpose was to improve the value of care compared with prevailing unstructured clinical judgment, which put select patients, as well as provider autonomy, at risk.

The eventual consensus criteria that the group defined for patients for whom GCSF prophylactic administration was warranted with each cycle of therapy were any one or more of the following:

• ECOG functional/performance status score of ≥ 2; 

• A prior episode of febrile neutropenia with either first- or second-line chemotherapy; 

• Prior evidence of prolonged neutropenia resulting in treatment delay during either first- or second-line chemotherapy; or

• Previous radiation accounting for > 25% bone marrow (BM) exposure (assessed through a BM distribution diagram).

Ultimately, in synthesizing all of the above considerations, the final consensus was a two-way course of action for a patient for whom any one or more of the above factors was true: it remained at the discretion of the provider to consider proceeding with standard dose cabazitaxel along with GCSF support or to consider dose reduction of cabazitaxel (without GCSF support). Per consensus, in such cases where the provider posits that there is no compromise of clinical benefit to the patient, the initial consideration will be for a one-level dose reduction of cabazitaxel.

Interestingly, the process of deliberation—based on the intuitive reliability and transparency of the predictor variables, the RF extrapolations, and the performance statistics—prompted the GUC providers to consider wider aspects and implications of their chemotherapy use. These included the commonly associated toxicity of diarrhea, their patients’ reported experiences, ongoing usage data, and overall goals of care with use of cabazitaxel.


In addition to orchestrating the development of the GCSF support rules and negotiating consensus, the CP architect had the challenge of arranging a process by which to extract performance data to evaluate provider compliance with the new rules. The initially proposed action was to embed a clinical prompt within the electronic medical record whenever cabazitaxel chemotherapy was ordered by a provider. A clinical prompt represented an obvious solution to the issue of integrating decision support rules into a clinician’s busy workflow. The limitation of a prompt is that it would function only to have a provider consider the patients’ oncologic history and even review the record for the details of the decision support rules defined above through consensus. While plausible, the lack of networked databases did not allow for extraction of all the elements of the consensus decision rules, specifically the structured details of radiation therapy exposure. Therefore, the current information silos did not justify the effort on the part of the IT department to embed a clinical prompt in the existing EMR.

Given this obstacle to clinical workflow adoption and dissemination of the rules, the necessary workaround was to use a repeated process of monthly GCSF use auditing, manual review of each patient record to determine compliance with consensus criteria, and monthly in-person reviews with each provider to positively reinforce proper implementation of criteria while ascertaining clinical reasoning for not applying criteria. The consensus criteria were also available to all providers, if needed, through an intranet link to the GMAP, which had been established as a primary resource for treatment selection and sequencing. In addition, a monthly operational report was provided to the GUC clinical director for review at monthly faculty meetings, with the intention of providing another prompt for continued vigilance on the part of the providers.


The post-implementation audits over the ensuing 6 months showed a clear benefit. Specifically, 90% of primary GCSF use in patients receiving cabazitaxel met criteria. Those cases that were not in compliance were able to be attributed to inability to access data elements that the provider needed at the time. Prospective use of cabazitaxel within the organization decreased within the context of other drug approvals and their prioritization over cabazitaxel.


Despite the multidimensional challenges associated with the development, orchestration, and implementation of CPs and decision support rules, the above GMAP experience highlights an organized effort to address care variation in the absence of definitive studies and data that answer nuanced clinical questions. Beyond the various limitations that could be pointed out regarding validity, sensibility, or otherwise, the fundamental success of this endeavor lay in it being recognized as a beginning rather than a means to an end. Efforts to operationalize value-based care using payment and financial pressures without dedicated practice reform will propagate inefficiencies within each organization and systematically across the health care sector.Highlighted by the GCSF example above, organizational efforts that begin to address incumbent practices through well-defined, individualized initiatives that seek to leverage existing data assets are necessary to inform and guide ongoing systems reform.

To date, organizational reforms have been characterized by reactive management and policy efforts addressing fragmented aspects of care delivery at the level of individual providers, clinics, or departments. As in the case of GCSF, the impetus of ad hoc responses to pay-for-performance requirements or avoidance of declined reimbursement due to a practice variation only functions to undermine efforts at value-based care. Within this context, the successes of the above decision support rule development process and subsequent challenges include its impact on driving the developing nimble data platforms and infrastructures for knowledge management, knowledge creation, and point-of-care translation.20

The implementation effort in Phase 2 of the GMAP initiative provided insight into the fact that value creation for an organization does not immediately translate into value capture. In other words, although we successfully outlined the framework for achieving the proposed value outcomes, actually executing and materially securing that proposed value proved more difficult.  By concentrating this intervention within the GUC disease center, the CP architect was able to identify the relevant stakeholders, achieve their buy-in and use the GMAP initiative to structure the GCSF criteria development process. Commitment and momentum were maintained by delineating the value proposition offered by GCSF decision support to each stakeholder, by establishing a strong tie to overarching organizational goals of value-based care, and through the behavioral incentive to prevent the loss of provider decision-making autonomy. In parallel, organizational cultural underpinnings pointed to the role of formal endorsements from GUC disease center leadership, emphasizing the significance of this effort. 

Evaluation of GCSF usage is ongoing across the different disease centers within the organization. Metrics of effectiveness include clinical outcomes of infections, hospitalizations, and chemotherapy treatment discontinuation. Besides GCSF use criteria, the generalizability of a rapid HTA approach to any practices or processes ultimately depends on two factors: ability to assess clinical value and infrastructure to capture that value.


The development and implementation of any new clinical decision support rules without the benefit of robust data and conventional methods such as RCTs remains a challenging undertaking. The successful application of a systematic HTA approach was accomplished through use of core ingredients of a value-based focus, orchestration by a clinical champion, identification of and engagement with key stakeholders, and respect for provider workflows. With the rapid pace of new data assets resulting in changing practices and treatment paradigms, institutions must evolve their HTA infrastructures and processes to meet the goals of value-based oncology care. 



1.     Siegel R, Ma J, Zou Z, Jemal A. Cancer statistics, 2014. CA Cancer J Clin. 2014;64(1):9-29.

2.     Porter ME. What is value in healthcare? N Engl J Med. 2010;363:2477-2481.

3.     Doebbeling BN, Chou AF, Tierney WM. Priorities and Strategies for the Implementation of Integrated Informatics and Communications Technology to Improve Evidence-Based Practice. J Gen Intern Med. 2006;21(supp 2):S50-S57.

4.     Blumenthal D, Tavenner M. The “meaningful use” regulation for electronic health records. N Engl J Med. 2010;363(6):501-504.

5.     Hailey D. A preliminary survey on the influence of rapid health technology assessments. Int J Technol Assess Health Care. 2009;25(3):415-418.

6.     Hassett MJ, Neville BA, Weeks JC. The relationship between cost, quality, and outcomes among women with breast cancer in SEER-Medicare [ASCO abstract 6001]. J Clin Oncol. 2011; 29(suppl):S384.

7.     D3 Oncology Solutions. D3 Oncology Solutions Web site. Accessed March 2, 2016.

8.     Panella M, Marchisio S, Di Stanislao F.  Reducing clinical variations with clinical pathways: do pathways work?  Int J Qual Health Care. 2003;15(6):509-521.

9.     Meropol NJ, Schrag D, Smith TJ, et al. American Society of Clinical Oncology guidance statement: the cost of cancer care. J Clin Oncol. 2009;27(23):3868-3874.

10.     Djulbegovic B, Loughran TP Jr, Hornung CA, et al. The quality of medical evidence in hematology-oncology. Am J Med. 1999;106(2):198-205.

11.     Vincent S, Djulbegovic B. Oncology treatment recommendations can be supported only by 1-2% of high-quality published evidence. Cancer Treat Rev. 2005;31(4):319-322.

12.     Hewitt ME, Simone JV eds. Ensuring Quality Cancer Care. Washington DC: National Academy Press, 1999.

13.     Schneider EC, Malin JL, Kahn KL, Emanuel EJ, Epstein AM. Developing a system to assess the quality of cancer care: ASCO’s national initiative on cancer care quality. J Clin Oncol. 2004;22(15):2985-2991.

14.     Malin JL, Schneider EC, Epstein AM, Adams J, Emanuel EJ, Kahn KL. Results of the National Initiative for Cancer Care Quality: how can we improve the quality of cancer care in the United States? J Clin Oncol. 2006;24(4):626-634.

15.     Khangura S, Konnyu K, Cushman R, Grimshaw J, Moher D. Evidence summaries: the evolution of a rapid review approach. Syst Rev. 2012;1:10.

16.     Smith TJ, Hillner BE. A way forward on the medically appropriate use of white cell growth factors. J Clin Oncol. 2012;30(14):1584-1587.

17.     Green SM, Schriger DL, Yealy DM. Methodologic standards for interpreting clinical decision rules in emergency medicine: 2014 update. Ann Emerg Med. 2014;64(3):286-291.

18.     Custer RL, Scarcella JA, Stewart BR. The Modified Delphi Technique - A Rotational Modification. J Vocational Technical Education. 1999;15(2):50-58.

19.     Watt A, Cameron A, Sturm L, et al. Rapid reviews versus full systematic reviews: an inventory of current methods and practice in health technology assessment. Int J Technol Assess Health Care. 2008;24(2):133-139.

20.     Elfiky AA, Zhang D, Krishnan Nair HK. Practice innovation: the need for nimble data platforms to implement precision oncology care. Discov Med. 2015;20(108):27-32.