Cancer pain is prevalent and long-term opioid therapy (LTOT) is commonly required.1 The global prevalence of cancer pain is 44.5%.2 Though therapeutically appropriate, when treating cancer pain that is unresponsive to simple analgesics with opioids, there are still risks of side effects, opioid tolerance and opioid misuse.3 These issues are particularly prevalent in the growing population of disease-free cancer survivors and patients living with chronic cancer, a risk that at times may be underestimated.4

This makes LTOT a challenging clinical decision requiring careful consideration of risks versus benefits.5 Providers consider multiple factors before prescribing opioids including pain severity, the underlying cause (eg, cancer or comorbidities), prognosis, history of substance use or mental illness and alternative treatment modalities.5 Since guidance and screening tools unique to patients with cancer have only recently been developed and are not routinely employed in the clinic, providers express a lack of confidence in the management of these patients.6–8 Moreover, these guidelines do not necessarily encompass many other non-disease-related sources of ‘non-clinical factors’ in decision-making.9

These non-clinical factors can be categorised into three categories: (1) Patient-related factors such as patient socioeconomic status (SES), race, age, attitude/behaviour, etc. (2) Provider-related factors such as personal characteristics, age, gender, culture, etc. (3) Practice-related factors such as availability of healthcare resources, practice type, size of practice, etc. Considering the significant, yet challenging decision of LTOT in patients with cancer, it is important to understand how patient, provider and practice-related factors influence this decision. In this secondary analysis of qualitative interviews with prescribers from the USA and Australia, we aim to identify these factors and understand their impact.

We undertook a secondary qualitative analysis of two research studies in the USA and Australia aiming to understand providers’ decision-making process when weighing the risks versus benefits of LTOT for patients with cancer. A more detailed explanation of data collection and interview guide are available in previous publications.5 10 Reporting adheres to the Consolidated Criteria for Reporting Qualitative Research.11

The US-based interviews were conducted in two Veterans Health Administration (VHA) healthcare systems (California and Connecticut) among both primary care and oncology providers. The interviews in Australia occurred mostly in New South Wales in primary care settings with general practitioners (GPs) involved in cancer pain management. This was due to the primary care providers’ large role in opioid prescribing.12

In each setting, the interviews were conducted by PhD-trained social scientists (KG and TL). The data collection happened concurrently. A semistructured interview guide was designed by the US-based team including KG, KL and WB.10 The interview guide was developed iteratively by a panel of experts in pain management, ethics, oncology, palliative medicine and qualitative methodology. The US guide was shared with the Australia-based team and questions were adapted for that context. The Australian interview guide was drafted with the help of members of the Consumer Advisory Panel of the Improving Palliative, Aged and Chronic Care through Clinical Research and Translation Centre at the University of Technology, Sydney. KL and KG were members of both research groups and were included in the protocol and IRB of both studies.

In the USA, we employed a quota sampling approach capturing the perspectives of 20 providers (10 primary care providers and 10 oncology-based personnel). In Australia, we employed a purposive sampling approach recruiting 22 GPs across the country. Interviews lasted between 20 and 60 min.

US providers were recruited via electronic email. Providers were based in two VHA centres on each US coast. Australian providers were recruited via email listservs, conferences/forums, phone, email and direct interaction.5 10 Using a quasi-randomised approach, we sampled from various regions with the aid of Google Maps.10 Australian participants were paid for their time at a standard GP rate.

The interviewers had no relationship with any of the participants before the interviews. We informed participants about the purpose of the study beforehand. The interviewer and interviewees were the only participants. All participation was voluntary with written informed consent. All interviews were audiorecorded and professionally transcribed and no interview was repeated.

Qualitative analyst SF and PhD-trained social scientist KG analysed the data analysis with input from clinician researchers: KL and AZ. We first extracted factors impacting provider decisions on LTOT in patients with cancer using Atlas.ti.13 There were 81 examples (51 from the USA and 30 from Australia). Next, we did an open and axial coding of the data set with a dual review to identify themes presented in both the USA and Australia (online supplemental table 1), ultimately collapsing codes into themes that represented all examples in the data set (online supplemental table 2).14

Although patients were interviewed as a part of the Service Directed Research (SDR) project, they were not involved with the study design. We have informed patients who there will be a publication, however, we will not be contacting participants individually after publication.

Our study illustrates the non-disease-related factors that influence the process of weighing the risks versus benefits of LTOT for cancer-related pain. Providers in our study enumerated patient personality type, mental health history, diversion risk via family, race/ethnicity, housing, SES and therapeutic relationship length factored into their decision-making. They further mentioned their personal experience, training type, and appointment time, and practice-related factors such as prescribing policies and availability of alternatives to pain management as other influencers.

When these factors predictably deviate from rationality in judgement or decision-making, they can be seen as ‘cognitive bias’.15 Provider bias is associated with inaccurate diagnosis and suboptimal medical management.16–18 Age, gender, years of practice and specialty of the providers are known factors that impact analgesic prescription.19–22 In the context of pain, patient race/ethnicity, SES, gender and English proficiency have been identified as risk factors for pain management disparities.23–25 There are two dynamics at play in a clinical interaction that could be impacted by bias. First is the clinician’s perception of the patient as a person, which is influenced by implicit and social biases based on the patient-related factors mentioned in the study. Second, the clinician heuristics and cognitive process of interpreting clinical data which are influenced by the provider-related factors mentioned by the interviewees.26

Provider cognitive biases can disproportionally impact the care of patients from disadvantaged backgrounds. Intersectionality describes the distinctive issues that arise when multiple identities of disadvantage (such as gender, class, ethnicity) intersect in the same person.27 Patient categories of identity such as race, SES, education, English fluency and mental illness can put patients at risk of discrimination in healthcare or pain management.23–25 28–30 For instance, providers in our study mentioned time availability as a factor that would impact their approach to pain management. When an individual is under time pressure, they are most likely to default to system 1 thinking because it is rapid, less taxing and relies heavily on heuristics or cognitive ‘shortcuts’.26 31 Indeed, during hectic clinical schedules providers spend less time understanding patient context and can have higher implicit bias and an increased tendency to stereotype minority patients.32–34 Providers in our study further cited negative past experiences with patients who misused opioids and existing therapeutic relationships with a patient as other impacting factors. In disadvantaged settings where providers do not know their patients well due to limited care continuity care and are faced with a higher burden of opioid misuse, it is unsurprising that these biases could intersect and impact the quality of pain management.

Providers may treat patients who live in low SES communities and have a high burden of opioid misuse, with a risk aversion bias. Sharing prescription medication is a common phenomenon in some communities.35–38 However, with opioids, there is an important risk of family members overdosing or developing substance use disorder.39 40 There is simultaneously a high rate of hazardous substance use and psychiatric illness in low-SES communities.41 42 As such, providers in our study may have been justified in their hesitation to prescribe opioids to patients who are low SES and have families that had an ongoing or history of opioid use. However, there is the danger of risk aversion bias and undertreatment of pain in low SES patients who may have these risk factors but are also capable of opioid safekeeping.17

Race is another well-established factor in implicit bias,26 though its intersection with other demographic factors such as SES and geographical location is less discussed. For instance, when providers considered both ethnicity and SES of patients, African Americans were less likely to be prescribed opioids compared with whites from the same SES.43 Similarly, an Australian provider in our study discussed Aboriginals may be at a higher risk of misuse in an urban setting compared with rural areas as there is greater continuity of care and a lower risk of doctor shopping in rural areas.

Regulator interventions may impact opioid prescription further exacerbating biases that impact the quality of care. For instance, being identified as the top 20% of opioid prescribers sometimes deterred Australian providers from prescribing in appropriate contexts. Although drug monitoring programmes potentially decrease the number of opioid prescriptions, it is unclear how they impact appropriate versus inappropriate prescribing.44 In fact, these initiatives may negatively impact the quality of care for patients from a minority background.45–47 For instance, in one US study, providers were more likely to discuss drug monitoring reports with Hispanic patients and discontinue opioid therapy for a black patient with a positive urine test compared with a white patient.45 47 Additionally, as compliance with these programmes is time-consuming, it limits available clinical time further fueling system 1 thinking.26 32–34

Practice-related factors are a part of this intersectionality. The providers in disadvantaged areas mentioned heavily relying on pain management with opioids due to the poor quality and inaccessibility of complementary and integrative health modalities. These therapies, based on the interdisciplinary biopsychosocial approach to pain management, are particularly important in patients from disadvantaged backgrounds due to their multimorbidity and complex psychosocial needs.48 49 These patients have worse pain outcomes due to healthcare inaccessibility, social support deficit, stressful and unsafe environments, and mental illness.43 50–52 Despite significant need and interest, complementary and integrative health modalities remain inaccessible.53 54

The patient-related factor, poor prognosis is often cited as the delineating factor between chronic pain in patients with and without cancer.55 However, with 18.1 million cancer survivors and a 69% chance of 5+ years survival, LTOT for cancer survivors carries the same risks as for patients with pain from non-cancer aetiologies.56 In non-cancer aetiologies, tools such as the pain medication questionnaire and the screener and opioid assessment have been developed to circumvent some of the mentioned biases and predict misuse.57 Building on this work with information theory and proper probability models, we can design clinical decision support tools that assess the risk of LTOT in all patients regardless of their cancer status.58 With the help of informatics, we can identify more accurate clinimetries to predict opioid risks, perhaps replacing broad categories such as race or cancer.59 These decision support tools can lower provider cognitive load, decrease bias and improve consistent, safe LTOT for both patients with pain from cancer and non-cancer aetiologies.60 61

This study can be considered in light of the following limitations. This was a secondary analysis of the data collected for two parent studies in Australia and the USA. Since the study was designed after data collection, the parent studies do not have exactly parallel samples (eg, the Australian sample includes GPs only and the US sample includes GPs and oncology-based personnel). The interviews did not specifically probe into the non-disease-related factors that influence clinical decisions; however, they all investigated how prescribers generally weigh the risks versus benefits of opioid prescribing. Reflection on non-disease-related factors and potential biases and consequent disparities emerged in the majority of interviews. All of the indicated themes were covered with various granularity in both datasets. A primary future quantitative survey study can help us confirm the factors mentioned by the provider. A supplementary qualitative study may help us uncover other factors not mentioned by our providers.