Case and management strategy development

Our team previously conducted two studies that provided insight into how to design cases and responses (clinical management strategies).16 17 In a recent qualitative study, we asked palliative care clinicians to describe challenges they face when caring for patients with serious illness prescribed opioid therapy. Clinical management challenges identified included managing opioids in patients with a history of substance use disorders and addressing opioid-related harms, including misuse behaviours such as concurrent use of cocaine or medications that were not prescribed such as benzodiazepines.16 In addition to substance use, other prior work has identified several common and challenging misuse behaviours that may arise in patients prescribed opioid therapy such as taking more opioids than prescribed or aggressive behaviour.17 Importantly, our qualitative study consistently identified prognosis for life expectancy (referred to in this article as prognosis) as an important factor that influences opioid decision-making in individuals with serious illness. For example, many participants thought that opioid-related harms may be a less important consideration when prognosis is short. We used these findings as a basis for writing Delphi study cases.

We used these prior studies as a starting point to create cases with challenging clinical characteristics.16 17 Additionally, given the importance of prognosis in decision-making, we decided to present the same cases and management strategies in a hypothetical patient with a prognosis of weeks to months, and separately in a patient with a prognosis of months to years.

All vignettes started with the same basic case with a prognosis of either weeks to months or months to years. The study team decided that this basic case should include attributes that (1) would lead many clinicians to prescribe opioids, (2) are common in ambulatory palliative care and (3) for which opioid therapy would be guideline-concordant care.18 19 Attributes within final cases described a middle-aged patient of unspecified gender or race, with advanced cancer, undergoing treatment and experiencing pain. This basic case was as follows:

You are seeing a 50-year-old patient with advanced cancer (defined as cancer that is unlikely to be cured or controlled with treatment). They are on active anti-cancer treatment. They have pain related to their cancer or its treatment. The patient’s prognosis is weeks-months [in second panel, months-years]. Assume that you have your X waiver to prescribe buprenorphine/naloxone for opioid use disorder (OUD) and that the patient’s insurance covers buprenorphine/naloxone if needed.

Table 1 provides the full text of the cases. The study team prioritised the development of cases that represent particularly common or challenging issues raised in our preliminary study16 and in our prior work on opioid misuse behaviours.17 Ultimately, we developed seven cases which each added additional clinical information beyond the basic case centering on the following management challenges: (1) inadequate pain control on highest recommended dose of buprenorphine/naloxone, (2) inadequate pain control on stable methadone dose, (3) requests early refills, (4) positive urine drug screen for benzodiazepines that were not prescribed, (5) positive urine screen for cocaine or methamphetamine, (6) aggressive patient behaviour in clinic and (7) history of untreated OUD not currently on pharmacologic treatment, with unmanaged pain.

Each case was followed by several questions on appropriateness of various management strategies based on published management strategies for opioid misuse behaviours in primary care settings20 and study team clinical expertise. These included strategies such as increasing opioids; tapering opioids; starting, splitting and stopping buprenorphine/naloxone, methadone, or other full agonist therapy and referring patients to addiction treatment. Appropriateness was queried using a 9-point Likert scale from the RAM,21 which ranges from ‘very inappropriate’ to ‘very appropriate’. Free-text boxes were also provided to allow participants to comment on additional information they would need to inform their management for each case, or provide other relevant thoughts.

The study team piloted these cases and management strategies using a cognitive interviewing-based approach, in which cases and responses were read aloud and assessed for clarity, understanding and content.22 Cases and strategies were iteratively refined and then finalised.

Recruitment

Participants were recruited online from the American Academy of Hospice and Palliative Medicine, Hospice and Palliative Nurses Association, Buprenorphine Clinician Support Network, Society of General Internal Medicine Pain and Addiction Shared Interest Groups, American Academy of Hospice and Palliative Medicine Addiction Shared Interest Group and the Palliative Care Research Cooperative Pain and Opioids Special Interest Group. Although membership in these groups is not limited geographically, these are all US-based groups. A list of additional experts to approach individually was generated by the study team and recommendations from potential participants.

Eligibility criteria

Potential participants were emailed a survey to determine eligibility. Participants were eligible to participate if they were >18 years old and (1) were board-certified in addiction medicine, palliative care or both; (2) had trained (in residency or fellowship) in addiction medicine, palliative care or both or (3) demonstrated other expertise in adult addiction or palliative care (were waivered to prescribe buprenorphine/naloxone for OUD; prescribe buprenorphine/naloxone, methadone or other opioids in palliative care or addiction settings to manage pain or addiction; conduct research related to opioid prescribing in palliative care settings or outpatient OUD treatment or have spoken at national conferences about these topics). Individuals who met eligibility criteria and were willing to participate were prompted to complete a demographics survey that included gender, race, ethnicity, age, expertise, clinical role, time since completion of terminal degree and state of practice.

Panel creation

To minimise participant burden and allow participants to focus on case scenarios that all specify the same prognosis, participants were randomly assigned to participate in either the ‘weeks to months’ or ‘months to years’ prognosis panel. During randomisation, we stratified by participant expertise in palliative care or addiction. Given fewer numbers of participants with addiction expertise than palliative care expertise, we categorised participants who had both addiction expertise and palliative care expertise as being in the addiction category. We also stratified by professional identity (physician and advanced practice provider such as nurse practitioner). This approach was taken to balance the type of expertise on each panel.

Sample size

Previous research recommends including 40–60 participants in online Delphi studies.23 Our goal was to include a minimum of 40 participants per panel. We aimed to recruit 60 participants per panel to account for attrition.

Data collection

Data for all three Delphi rounds were collected using ExpertLens, a web-based platform developed by RAND that allows for participation in Delphi panels online.24

Round 1 began on 10 August 2020. Participants were asked to rate and comment on the appropriateness of management choices for the seven cases.

Round 2 began on 10 September 2020. The purpose of round 2 was to allow participants to consider other points of view and re-consider their round 1 responses. Participants were asked to review their own and other participants’ numerical responses and free-text comments from round 1. For each case and each management strategy, information was provided as to whether consensus was reached, and if consensus was reached whether the strategy was found to be appropriate, not appropriate or of uncertain appropriateness, based on the prespecified analytic approach (see the Data analysis section). Participants also viewed summaries of round 1 free-text comments. Participants, identified by anonymous ID numbers, then participated in asynchronous online discussion moderated by the study principal investigator (JM) and RAND co-investigator (DK).

Round 3 began on 17 September 2020. Participants again have the opportunity to review their own and other participants’ numerical and free-text comments/discussion from rounds 1 and 2. They are then given an opportunity to change their final numerical responses.

Data analysis

ExpertLens automatically identifies whether there is consensus, or disagreement, among responses in panels based on decision rules derived a priori using the RAM.21 25 26 Specifically, this method uses a two-step analytic approach: first, it identifies disagreement by evaluating the distribution of ratings. If no disagreement exists, it uses the median value to determine whether the panel rating was positive, negative or uncertain (figure 2).21 We used this automatic process to analyse data from rounds 1 and 3. As round 2 is purely discussion-based, there was no analysis plan. Of note, only round 3 responses will be used to assess final consensus and disagreement. Qualitatively, we analysed free-text comments from all rounds using thematic analysis.27 28 We grouped comments for each strategy by numeric ratings to which they referred. Data were coded by three individuals trained in qualitative analysis and supervised by an expert in these methods (DK) who reviewed all results. Collectively, the team met to discuss coding disagreements until consensus was reached. The lead author (JM) reviewed final codes to ensure correct interpretation of qualitative data clinically.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Statistical approach to analysing data about appropriateness of management strategies from a modified Delphi panel.25 30

Patient and public involvement

No patients involved.