AML Clinical Trial Decision Making: How Often Do Physicians Follow a Consensus-Derived Assignment Scheme?

Background

At our center we consider most patients with newly-diagnosed AML or MDS-EB2 to be candidates for clinical trials. The NCI website (clinicaltrials.gov) currently lists 72 adult trials implying uncertainty as to which is best. These trials often have arbitrary exclusion criteria, and even if eligible, not every patient may enroll in a given trial. Other sources of clinical trial selection bias include physician or patient preferences and logistical issues such as transportation or insurance coverage. These selection biases may contribute to the well-known difficulties in reproducing "promising" findings from early phase trials in subsequent larger studies. At our center we use a pre-determined treatment assignment scheme to allocate clinical trials to patients prior to the initial physician-patient consultation, which may increase trial enrollment and decrease bias in trial assignment. To assess selection bias, we examined how often the assigned treatment was given to patients.

Methods

A randomly-ordered, physician consensus-derived clinical trial list was created and updated as protocols opened and closed. Trials initiated by our investigators appeared more often on the list than trials sponsored by industry partners. Upon a patient's arrival we provided her/his physician a pre-assigned clinical trial based on that patient's age and probability of early death (using the treatment related mortality score, TRM). After being informed of options including other trials, standard care, or supportive care alone, physicians and their patients decided on a treatment. Once a treatment choice was selected, physicians received an electronic survey. In cases where the pre-assigned clinical trial was not chosen, the survey inquired about reasons for this decision. Categories were: (a) patient ineligible, (b) protocol not accruing, (c) patient and/or physician preference and (d) logistics, e.g. patient cannot stay in Seattle for length demanded by protocol) or (e) insurance denial.

Results

From December 2017 through June 2019, we collected surveys on 158 out of 170 (93%) arrivals for whom treatment decisions were made.

The pre-assigned therapy was given in 40/158 cases (25%, 95% confidence interval [CI] 19-33%). In all 40 cases the trial chosen were ones initiated by investigators here. Twenty-four of the 118 patients (21%) in whom the assigned therapy was not given chose to be treated at home or receive palliative care only. Reasons the remaining 94 patients did not receive their assigned therapy were: physician/patient preference (51), protocol not accruing (18), logistical issues (13), and ineligibility (12 patients). Among patients not enrolled on the assigned trial due to physician/patient preference, 27 of 51 received treatment on an alternate clinical trial, and 24 received standard therapy. The intensity of the therapy was identified by the treating physician as a factor leading to the alternate treatment choice in 35 of 51 patients (67%). In 21 patients in whom a less intense therapy was assigned, intense therapy was given instead; while the opposite occurred in 14 patients. In 16 cases the preferred intensity was the same as that assigned but a different regimen was used. The median TRM score was lower among patients given high intensity vs. intermediate/low intensity treatments (medians 3.75 for high vs, 10.05 for low; Kruskal Wallis p < 0.00001).

Discussion and conclusion

The pre-assigned therapy was given to 25% of patients (40/158), rising to only 31% (40/128) after removing cases where the protocol was not enrolling or the patient ineligible. Factors possibly contributing to these low rates are the number, nature and specifics of the available protocols, the characteristics of the patients referred to us (e.g. distance from Seattle to their homes), availability of acceptable alternative treatment in a local setting, and characteristics of our physicians (e.g. inherent preferences for higher or lower intensity therapy).

The low rate with which pre-assigned treatments were given suggests the continued presence of considerable selection bias when treatment decisions are made. The possibility of such bias should be kept in mind when interpreting the results of clinical trials. Improvements in care of AML patients will require trials to enroll patients representative of the population in need of treatment.