A scoping review of practice recommendations for clinicians’ communication of uncertainty

Abstract Background Health‐care providers increasingly have to discuss uncertainty with patients. Awareness of uncertainty can affect patients variably, depending on how it is communicated. To date, no overview existed for health‐care professionals on how to discuss uncertainty. Objective To generate an overview of available recommendations on how to communicate uncertainty with patients during clinical encounters. Search strategy A scoping review was conducted. Four databases were searched following the PRISMA‐ScR statement. Independent screening by two researchers was performed of titles and abstracts, and subsequently full texts. Inclusion criteria Any (non‐)empirical papers were included describing recommendations for any health‐care provider on how to orally communicate uncertainty to patients. Data extraction Data on provided recommendations and their characteristics (eg, target group and strength of evidence base) were extracted. Recommendations were narratively synthesized into a comprehensible overview for clinical practice. Results Forty‐seven publications were included. Recommendations were based on empirical findings in 23 publications. After narrative synthesis, 13 recommendations emerged pertaining to three overarching goals: (a) preparing for the discussion of uncertainty, (b) informing patients about uncertainty and (c) helping patients deal with uncertainty. Discussion and conclusions A variety of recommendations on how to orally communicate uncertainty are available, but most lack an evidence base. More substantial research is needed to assess the effects of the suggested communicative approaches. Until then, health‐care providers may use our overview of communication strategies as a toolbox to optimize communication about uncertainty with patients. Patient or public contribution Results were presented to stakeholders (physicians) to check and improve their practical applicability.


| INTRODUC TI ON
Uncertainty is pervasive in medicine. It has been defined as a metacognition-a self-reflective mental state in which one is subjectively aware of one's ignorance. 1 Rapid technological developments have yielded not only vast amounts of new biomedical knowledge but also information that may be difficult to interpret and/or overly complex. 2 Moreover, the rise of evidence-based medicine has paradoxically increased collective awareness within medicine of what is still unknown. [3][4][5] For example, the meaning and implications of genetic mutations cannot always be known, it is unpredictable which treatment will best benefit individual patients, and diagnostic test results may be difficult to interpret. 5,6 Health-care providers, patients and health researchers increasingly have to deal with these uncertainties. 7 In parallel with growing awareness of uncertainty, patients' roles have shifted: their information rights are being increasingly formally acknowledged. Moreover, there has been a rise in shared decision making, whereby health-care providers are expected to involve patients in decisions about their health and treatment. 8,9 However,to properly inform patients and justify their autonomy, clinicians need to be fully open about what they do and do not know. [10][11][12][13][14] In practice, this means they have to share with patients different types of uncertainty.
Both theory and research on uncertainty have been expanding in the past decades. The many manifestations of uncertainty have been approached from various disciplines, ranging from economics and mathematics to philosophy, psychology and sociology, which has resulted in wide variability in conceptual models and terminology. 1,5 Within health care, two main types of uncertainty are generally distinguished, albeit using various labels. First, first-order uncertainty, probability or aleatory uncertainty refers to the inability to predict future outcomes regarding, for example prognosis, or treatment effects, and often involves using risk estimates. 1 Second-order or epistemic uncertainty can arise either from ambiguity or from complexity of information, and may for example concern the inability to interpret test results or to provide a definitive diagnosis. 15 In 2011, Han 1 proposed a comprehensive taxonomy of uncertainty in health care, distinguishing not only various types-or causes-of uncertainty, but also additionally different issues to which uncertainty may pertain (eg, scientific issues regarding diagnosis, prognosis, cause and treatment of a given medical condition, as well as practical and personal issues), and the locus-or person(s)-in whom the uncertainty resides.
Most empirical research on communicating uncertainty to patients has focused on effects and implications of risk communication (first-order uncertainty). 5 It has yielded specific recommendations on how to convey risk information, particularly in written form. 5,[16][17][18] Yet, despite the significance of written information (eg, writing or drawing, information leaflets and websites), clinicians' oral information provision during medical encounters is considered by patients as their most important information source. [19][20][21] Research on interpersonal oral communication, specifically regarding second-order uncertainty, is more scarce, and its results have been inconclusive. 22 For example, studies have reported contradicting effects of physicians' uncertainty expressions on patient satisfaction. [23][24][25][26]27 These conflicting results may be explained by variation across health-care providers' communicative approaches to conveying uncertainty. Physicians' expressions of uncertainty were found to be detrimental to patient satisfaction particularly if physicians did not perform actions to support patients in managing the uncertainty. 25 In another study, clinicians' explicit acknowledgement of uncertainty (eg, "I don't know") was detrimental to patient confidence, whereas behaviours implying uncertainty (consulting a book or colleague) were seen as benign or even beneficial to trust. 28 Apparently, discussing uncertainty with patients can have variable effects, depending on health-care providers' specific communication approaches, and providers may need to tailor their communication strategies according to the specific uncertainties at hand.
Despite general agreement that providers should discuss uncertainty with patients, to our knowledge no comprehensive guidelines are available for how to do so. This is problematic, as it could result in unwarranted practice variation in provider-patient communication.
As a result, patients might be exposed to suboptimal communication, inducing possible underrecognition, excessive awareness and/ or misunderstanding of uncertainty. 16 Eventually, this could result in feelings of uncertainty and anxiety, misunderstanding, impaired decision making and/or reduced satisfaction with and trust in their physician. 29 We sought to create an initial overview of the practical recommendations on how health-care providers can orally communicate uncertainty to patients within clinician-patient encounters.
We focused on second-order uncertainty rather than publications focusing exclusively on first-order uncertainty. Using a scoping review of the empirical and non-empirical literature, we identified which practical recommendations are available for health-care providers to discuss uncertainty with patients, and how evidence-based these are.

| ME THODS
A scoping review was deemed the most appropriate type of review to meet our aim, as it enables exploring the breadth of existing research, comprehensively mapping the literature and providing directions for future research. 30,31 We developed the review protocol K E Y W O R D S clinicians, communication, disclosure, health care, health personnel, physician-patient relations, referral and consultation, review, uncertainty using the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR). 32

| Search strategy
A search strategy was set up in MEDLINE and then translated to CINAHL, EMBASE and PsycINFO (see Table 1). It included any variations of the following keywords: (1) clinical practice guideline, (2) communication, (3) uncertainty and ( 4) health-care providers. Note that "risk" and "risk communication" were deliberately not included in our search. We aimed to exclude the body of literature exclusively focused on risk as it often focuses on the more technical aspects of communicating statistical/numerical information (eg, using percentages vs frequencies). We searched databases from inception until 24 July 2019. Figure 1 illustrates the article selection process. Three reviewers (EvB, NM and MH) screened all titles and abstracts for eligibility, using Rayyan, 33 in three steps: first, 50 abstracts were screened jointly to further specify inclusion and exclusion rules; second, all three reviewers independently screened 200 abstracts and solved any discrepancies; and third, the remaining abstracts were independently screened by two reviewers each. Discrepancies were solved through discussion. Any types of English/Dutch-written abstracts (including dissertation abstracts, non-empirical papers) were included if describing any type of recommendation-evidence-based or not-for any type of health-care provider on how to (not whether to) orally communicate uncertainty to patients. Articles were excluded if they (1) included only an abstract (eg, conference proceedings); (2) included only recommendations for providers to help patients deal with uncertainty; and (3) concerned only risk communication in its narrowest sense.

| Article selection and exclusion criteria
Full-text screening was performed by two (out of three) reviewers independently, using the same criteria. Any discrepancies were discussed and resolved during consensus meetings. Reference lists of all papers included after full-text screening were checked for additional relevant papers.

| Data extraction
Two reviewers (NM and MH) performed data extraction using an instrument based on the PRISMA and RAMESES guidelines. 34,35 Data from the first three articles were extracted jointly to fine-tune the extraction form. The remaining data were extracted by one reviewer each, and any doubts were discussed. Aside from descriptive characteristics for each study (eg, year of publication, publication type, design and population), we extracted any recommendations regarding the communication of uncertainty that were provided. For these recommendations, we specified (1) medical setting and context/topic to which the recommendations applied (eg, diagnosis, treatment decision); (2) target group, that is for which health-care providers the recommendations were intended; and (3) strength of the evidence base (ie, whether findings were based on new or previous empirical findings, non-empirical literature or no evidence). Assessment of these criteria was inductive, meaning that no pre-specified criteria 1 Clinical practice guideline [MeSH] 2 (recommend* or advice* or advis* or tips or suggestion* or strategy or strategies or approach* or practice or principle* or skills or training or problems or (clinical adj1 practice adj1 (variation or pattern))).AB,TI,KF (disclos* or communicat* or discuss* or conversat* or interact* or explain* or explanat* or ((provision or provid* or disseminat* or convey* or deliver* or exchang*) adj3 (information or result* or outcome or message*))).AB,TI,KF

13
AND/3, 6,9,12 TA B L E 1 Search strategy used in MEDLINE were used, but rather these were allowed to emerge from the data. 36 No quality assessment of the empirical studies was performed, as the aim was to provide a scoping overview rather than a systematic appraisal.

| Narrative synthesis
After completing data extraction, the two reviewers reviewed and rephrased and/or summarized the recommendations, to enable comparison and aggregation of overlapping recommendations. A first draft of a list with recommendations was created and modified after review and discussion between authors (NM, AS, ES and MH).
Moreover, the list was discussed with two additional authors-one a general internist/palliative care physician (PH) and the other a clinical geneticist (CA)-to check and improve their practical applicability. Finally, the list was finalized (see Table 3).  Gerodontologists c When asking informed consent, prepare patients for disappointment and provide the opportunity to adapt to and cope with the unexpected. Do so by using stories of concrete, emotionally interesting information (ie, narrative) rather than through factual assertions of abstract data or statistical evidence;

| Characteristics of publications
and by segmenting problems, identifying choices and clarifying probabilities Maguire, 1988  Column indicates for all empirical studies whether the recommendations were directly derived from the study findings (eg, when different communication strategies were observed or compared), or indirectly based on study findings (eg, when clinical geneticists' views on uncertainty were explored).
b n/p means that the information was not provided by the publication.
c Information was not specified in the article but derived from other, additional information mentioned in the article.

Overarching aim Recommendation Explanation and examples
Preparing for the discussion of uncertainty

Warn patients for the possibility of uncertain outcomes
Before initiating treatment or diagnostic/genetic testing, prepare patients that they may be confronted with uncertain outcomes. [37][38][39][40][41] This should help patients manage their expectations, 42 and prepare them for continued uncertainty, 43  Be aware that for some patients and depending on circumstances, awareness of uncertainty may lead to negative emotions (eg, anxiety, feelings of uncertainty) and/or a more negative appraisal of their care provider. This strategy requires taking into account patient preferences and psychological capacity to tolerate uncertainty and adjusting one's communication accordingly. For example, when sensing that patients react negatively to explicit statements of uncertainty such as "I don't know", it may be better to resort to more implicit means that effectively reflect the complex reality. Examples are as follows: explaining the most likely diagnoses, 61 asking other doctors for advice 28 and using careful terms such as "maybe" 61,62 "It is possible that the test shows you have this disease" 5. Outline all potential scenarios and discuss their implications for patients' life Based on your expertise and the available knowledge, make predictions or draw preliminary conclusions. 58,59,63 Specifically, discuss a discrete set of potential diagnostic/prognostic scenarios or treatment options. 64 For prognostic communication, this is best done by providing best case, worst case and most likely scenarios, which may facilitate a sense of hope. 51,65 Take the patient along in your reasoning to enhance insight and knowledge. 63 Outline the potential impact of each scenario on the patient's life 58 "We cannot predict how your condition will develop, but there are roughly three scenarios: scenario 1 …" "In the worst case scenario […], in the best scenario […]. But the most probable scenario is […]" 6. Explain uncertainty in an understandable, concrete and structured way Explain uncertainty using understandable language, possibly even using analogies to facilitate understanding. 66 Use concrete and emotionally engaging narratives instead of factual and abstract assertions. 44 Provide structure by summarizing key points and slowly pacing the information. 37,40,67,68 Minimize conflicting advice and any form of vagueness 48,67 "I will tell you only the most important things right now. If you need more details, please feel free to ask me" Non-empirical publications included conceptual papers (n = 5), opinion papers (n = 5), reviews (n = 5), practical recommendations (n = 4), Show commitment by appearing concerned with the patient and confirming your continued support and availability throughout the process 25,47,56,72,73 "Even though we are limited in our possibilities to help your husband fight the disease, we will remain highly involved in his care" one review/case study and one comment. If the medical setting was specified (32/47 publications), it most commonly involved oncology (n = 8), clinical genetics (n = 7) and paediatrics (n = 3). In 25 articles, recommendations were targeted at "physicians" in general, whereas other publications addressed particular disciplines or health-care providers.

| Characteristics of recommendations
Recommendations for communicating uncertainty were explicitly stated to apply to the following context: communicating prognosis (n = 12), (shared) decision making (n = 12), conveying diagnosis (n = 6) or genetic/genomic testing (n = 6). Seven recommendations applied to discussing uncertainty in general. In four publications, the context of the recommendation(s) was not explicitly stated but inferred by the authors from other information in the article (eg, setting). In 23 publications, evidence directly obtained from empirical work reported in the paper led to the provided recommendations.
Thirteen of these could be considered "empirically supported", that is directly supported by study findings, for example when different strategies for conveying uncertainty were observed or compared.
Ten were "empirically implied", that is more indirectly deduced from the study results, for example when communication preferences were identified in interviews with clinicians. In the remaining 24 publications, recommendations were based on previous (non-)empirical literature (n = 19), or not substantiated by any literature (n = 5).

| Narrative synthesis of recommendations
From our narrative synthesis, 13 recommendations emerged addressing three overarching goals: (a) preparing for the discussion of uncertainty, (b) informing patients about uncertainty and (c) helping patients deal with uncertainty (see Table 3). Below, we discuss all recommendations, and which data acquired in empirical research reported in the publication itself supports them.

| Preparing for the discussion of uncertainty
We identified two recommendations aimed at preparing for the discussion of uncertainty. Eight publications included the suggestion that (1) providers should warn patients for the possibility of uncertain outcomes. [37][38][39][40][41][42][43][44] This recommendation applied particularly to situations involving future uncertainty, that is involving diagnostic or genetic testing. Only two publications described empirical studies directly supporting this recommendation, both reporting that patients appreciated if they were prepared for potential uncertainty resulting from genetic testing. 38,42 The recommendation to (2) explore patients' individual preferences, beliefs and coping styles regarding uncertainty and to adapt communication accordingly was (re)iterated in 11 publications. 5, [45][46][47][48][49][50][51][52][53][54] In two cases, this recommendation was empirically supported. One observational study found that cancer patients with a more active problem-solving coping style preferred receiving explicit prognostic information from their physician. 51 In a second observational study, women at risk of pre-term birth varied in their preferences for receiving additional information to reduce uncertainty. 48

| Informing patients about uncertainty
We distinguished six recommendations aimed at conveying uncertainty to patients. Eleven publications included the advice to (3) openly acknowledge inherent uncertainty and explain the degree and nature of the evidence that is available. 43,46,47,51,53,[55][56][57][58][59][60] This may be viewed as a "double-barreled" recommendation: clinicians may or may not add an explanation about the level of evidence to their acknowledgement of uncertainty. This recommendation particularly applied to diagnostic or prognostic settings and was supported by empirical data from three studies. In the first, caretakers of disabled older patients reported to prefer if uncertainty was openly discussed with them. 59 The second concluded, based on qualitative observations, that people receiving genetic counselling were able to actively participate in handling an uncertain diagnosis. 58 The third, experimental, study found that cancer patients had a more realistic understanding of the variation in survival time if possible variation in prognostic expectancies was explicitly discussed. 57 Three publications suggested that (4) providers should allow flexibility in the extent to which uncertainty is communicated, depending on the individual and circumstance. This strategy requires taking into account patient preferences and psychological capacity to tolerate uncertainty and adjusting one's communication accordingly, for example by using more implicit wording (eg, "it could be" rather than "I don't know"). 28,61,62 All three publications provided empirical support for this recommendation. Both a survey and an experimental study found that physicians' explicit (vs implicit) expressions of uncertainty reduced patient-perceived technical competence, trust, confidence and patient adherence. 28,61 A third (observational) study reported that general practitioners' implicit expressions of uncertainty did not affect patients' anxiety. 62 The advice to (5) outline all potential scenarios and discuss their implications for patients' life was put forth in six publications. 51,58,59,[63][64][65] This recommendation applied mainly to the diagnostic phase, that is when a definitive diagnosis cannot yet be made or when discussing the potential outcomes of diagnostic testing. Two observational studies provided empirical support. In one, oncologists were observed to discuss possible scenarios of diagnostic test outcomes, thus providing patients with insight into their reasoning. 63 In the second study, cancer patients reported to prefer prognostic information to be framed in terms of a best case, worst case and most likely scenario, compared with framings not including a wide range. 51 In total, seven publications included recommendations to (6)  In seven publications, it was suggested that clinicians should (10) provide patients and their close ones with some sense of control to counterbalance the uncertainty, particularly when conveying diagnosis or prognosis. 47,51,53,59,64,69,70 This may involve highlighting that while one cannot control circumstances, one can do one's best in any scenario to choose the optimal path forward. Evidence on which interventions might be most effective in providing control is lacking, but two observational studies proposed potential interventions. Two observational studies showed that both caretakers of disabled older patients and paediatric residents saw the benefits of emphasizing the controllable elements of an uncertain situation. 59,69 Seven publications suggested to (11) provide a sense of hope to patients and/or their close ones, for example, by alternating uncertain bad news with certain good news, particularly in serious illness. 50,[57][58][59][63][64][65] Two empirical studies provided direct support. In one, caretakers of disabled elders suggested that physicians should help them maintain a sense of hope to deal with uncertainty. 59 In the second, oncologists were observed to communicate uncertainty by continually alternating between uncertain (possibly) bad news and good, reassuring news-although no evidence was provided to validate this approach. 63 Ten publications included the suggestion to (12) facilitate patients' emotional responses to the uncertainty, and/or provide emotional support. 40,46,47,56,58,64,[68][69][70][71] This recommendation was not empirically supported. Lastly, the advice for providers to (13) emphasize their continued involvement with the patient's care was (re)iterated in five publications, although it was not empirically supported. 25 Table 2). Possibly, this reflects a heightened awareness of the many complex uncertainties inherent to these settings. However, uncertainty is highly prevalent in almost any clinical setting and many of the provided recommendations may therefore be applicable more widely. whereas on the other hand they were recommended to adjust the extent to which they convey uncertainty to individual patients, sometimes using more implicit rather than explicit wording (eg, "it is possible that…" instead of "I don't know"). This discrepancy may reflect the conflict between the normative goals of maximizing patient autonomy, that is informing about the limits of knowledge, vs enhancing patient well-being, that is less explicit emphasis on potentially threatening uncertainties. 5,77,78 Additionally, health-care providers may need to tailor their communication strategies to their own feelings and behaviours in response to uncertainty, which may vary widely. 79 Hence, certain strategies may work better for some than for others. For example, providers with lower tolerance for uncertainty may find particular benefit in explaining to patients the causes of uncertainty, emphasizing what is certain and/or making clear follow-up plans.
Alternatively, providers may need to be supported in tolerating uncertainty, which could eventually help them improve their clinical care.
Strengths of this review are, first, that we included both empirical and non-empirical literature from a wide range of settings in our synthesis, and, second, our systematic search, selection and extraction, which enhances reliable and generalizable results. Third, by presenting the evidence base for all recommendations, readers can assess the underlying evidence for the available recommendations.
Limitations are, first, that we may have excluded literature providing practical recommendations in the main manuscript only, and not in the title or abstract. Second, our synthesis may involve some degree of subjectivity. However, we attained maximal objectivity by continuously discussing our preliminary results within the full research group. Third, although we deliberately ignored the specific risk communication literature, recommendations from that area may be compared with the ones presented here to examine the extent of overlap and/or conflict.
The strategies for communication about uncertainty identified in this review align with and build on previous work, 47

ACK N OWLED G EM ENT
The authors would like to thank Eline van Bree (EvB) for her help with title and abstract, and full-text screening.

CO N FLI C T O F I NTE R E S T
None.

AUTH O R S ' CO NTR I B UTI O N S
Niki Medendorp: conceptualized the study, designed methodology, involved in formal analysis, investigated the data, curated the data, wrote the original draft, wrote, reviewed and edited the manuscript, visualized the data and administered the project.
Cora Aalfs: conceptualized the study, and wrote, reviewed and edited the manuscript. Anne Stiggelbout: conceptualized the study, designed methodology, investigated the data, wrote, reviewed and edited the manuscript, and supervised the data. Paul Han: conceptualized the study, and wrote, reviewed and edited the manuscript.
Ellen Smets: conceptualized the study, designed methodology, wrote, reviewed and edited the manuscript, supervised the data and acquired funding. Marij Hillen: conceptualized the study, designed methodology, involved in formal analysis, investigated the data, wrote the original draft, wrote, reviewed and edited the manuscript, visualized the data, supervised the data, administered the project and acquired funding.