Journal of Mental Health
ISSN: 0963-8237 (Print) 1360-0567 (Online) Journal homepage: http://www.tandfonline.com/loi/ijmh20
To admit or not to admit? The effect of framing on risk assessment decision making in psychiatrists Kiri Jefferies-Sewell, Shivani Sharma, Tim M. Gale, Chris J. Hawley, George J. Georgiou & Keith R. Laws To cite this article: Kiri Jefferies-Sewell, Shivani Sharma, Tim M. Gale, Chris J. Hawley, George J. Georgiou & Keith R. Laws (2015) To admit or not to admit? The effect of framing on risk assessment decision making in psychiatrists, Journal of Mental Health, 24:1, 20-23 To link to this article: http://dx.doi.org/10.3109/09638237.2014.951477
Published online: 04 Sep 2014.
Submit your article to this journal
Article views: 134
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=ijmh20 Download by: [University of California, San Diego]
Date: 05 November 2015, At: 17:42
http://informahealthcare.com/jmh ISSN: 0963-8237 (print), 1360-0567 (electronic) J Ment Health, 2015; 24(1): 20–23 ! 2015 Shadowfax Publishing and Informa UK Limited. DOI: 10.3109/09638237.2014.951477
ORIGINAL ARTICLE
To admit or not to admit? The effect of framing on risk assessment decision making in psychiatrists Kiri Jefferies-Sewell1,2, Shivani Sharma2, Tim M. Gale1,2, Chris J. Hawley2, George J. Georgiou2, and Keith R. Laws2 1
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
Hertfordshire Partnership University NHS Foundation Trust, Mental Health Unit, QEII Hospital, Howlands, Welwyn Garden City, Hertfordshire, UK and 2Department of Psychology, School of Life and Medical Sciences, University of Hertfordshire, Hatfield, Hertfordshire, UK Abstract
Keywords
Background: The way that information is presented is well known to induce a range of biases in human decision tasks. Little research exists on framing effects in psychiatric decision making, but it is reasonable to assume that psychiatrists are not immune and, if so, there may be implications for the welfare of patients, staff and the general public. Aims: To investigate whether presentation of risk information in different formats (frequency, percentage and semantic) influences inpatient admission decisions by psychiatrists. Methods: Six-hundred seventy-eight general adult psychiatrists read a short clinical vignette presenting a case scenario of a patient presenting for inpatient admission. One of four condition questions followed the vignette, incorporating either numerical or percentage probabilities and the semantic labels ‘‘high’’ and ‘‘low’’ risk. In each condition, the actual risk was identical, but the way it was presented varied. The decision to admit the patient or not was recorded and compared across conditions. Results: More individuals chose to admit the patient when risk information was presented in numerical form (X2 ¼ 7.43, p ¼ 0.006) and with the semantic label ‘‘high’’ (X2 ¼ 7.27, p ¼ 0.007). Conclusions: Presentation of risk information may influence decision making in psychiatrists. This has important implications for mental health clinical practice where clinicians are required to interpret probabilistic information within their daily work.
Admissions, decision making, framing, psychiatry
Introduction Mental health professionals (MHPs) make important patient-related decisions on a daily basis, and so a clear understanding of their decision-making processes and biases is imperative to achieving best practice. Research has shown that, in general, human decision-making can be manipulated toward both risk aversion, and risk seeking, when information about risk factors is presented in different ways (Tversky & Kahneman, 1981), and this bias has been shown to extend to medical settings (Marteau, 1989; McNeil et al., 1982). This suggests that decisions made by health professionals are not independent of the information presentation format, a fact that may have implications for patient care. More recent findings suggest that framing effects appear only when the subject matter has real value for the decision maker (Bloomfield, 2006), and the direction of the effect is dependent on the baseline level of risk presented as well as the decision context (Wang et al., 2001). Yamagishi (1997) illustrated that manipulation of base-rate information and Correspondence: Kiri Jefferies-Sewell, Learning and Development Centre, Hertfordshire Partnership University NHS Foundation Trust, The Colonnades, Beaconsfield Close, Hatfield, AL10 8YD, UK. Tel: 01707 253 834. E-mail: [email protected]
History Received 26 September 2013 Revised 18 February 2014 Accepted 27 May 2014 Published online 4 September 2014
anchoring points influenced rankings of ‘‘risk of death’’. The author showed, in two experiments, that participants judged the probability 1286 out of 10 000 (12.86% chance) more risky than 24.14 in 100 (24.14% chance). The aim of this study was to investigate what influence the presentation of risk information has on inpatient admission decisions by psychiatrists. Specifically, we explored the manipulation of numerical (frequency per 10 000) vs. percentage data and the impact of using the semantic labels of ‘‘high’’ vs. ‘‘low’’ risk admission decisions. Based on previous work, we predict that presenting risk information numerically (as opposed to percentages) and using the term ‘‘high risk’’ will produce the highest number of admission decisions.
Methods Six-hundred seventy-eight (398 male) general adult psychiatrists recruited from the Royal College of Psychiatrists (RCPsych) and Hertfordshire Partnership University NHS Foundation Trust (HPFT) participated in this study. Psychiatrists from HPFT would also have had membership with RCPsych, but this information was not requested, and recruitment was sought from both the Trust and the RCP. However, comparison of anonymous collector identification
Effect of framing on admission decisions
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
DOI: 10.3109/09638237.2014.951477
numbers confirmed that none of the participants completed the questionnaire more than once. Using an independent measures design, all participants were randomly assigned to one of four conditions and e-mailed the relevant link. The presentation of risk data was manipulated to be presented as either a frequency per 10 000 or as a percentage. Semantic labels of ‘‘high’’ and ‘‘low’’ were also inserted to indicate an expert’s opinion about the level of risk. Prior to this experiment, the questionnaire was piloted on eight consultant and junior psychiatrists during a routine teaching session. The rationale behind this phase of the study was to ensure that decisions within conditions were not polarised. The ‘‘numerical high’’ condition (see procedure section for explanation of each condition) was presented to all participants, and results showed that 50% of participants chose to admit. Psychiatrists accessed an internet link to the online site where they were able to complete the questionnaire at their convenience. After entering demographic information (age, sex, profession, area of profession and length of time in role), participants then read the vignette (same for all participants) describing a patient who presents as follows: The patient is a 33-year-old male. At the age of 18 a firm diagnosis of schizophrenia, paranoid type, was made. The clinical course over the past 15 years has included readmission, some suicidal behaviour and some threatening behaviour to others. He presents to you with paranoia needing prompt treatment attention. Imagine you have the benefit of an extract from a report made by a previous treating Unit that had known the patient for a long time and has sophisticated, and highly accurate, techniques for assessing key risks. This text was followed by one of the following four conditions, to which participants were randomly assigned: (A) ‘‘If one considered 10 000 similar cases, 70 would kill someone within a month if not admitted to hospital – a high risk’’. [Numerical/high] (B) ‘‘If one considered 10 000 similar cases, 0.7% would kill someone within a month if not admitted to hospital – a high risk’’. [Percentage/high] (C) ‘‘If one considered 10 000 similar cases, 70 would kill someone within a month if not admitted to hospital – a low risk’’. [Numerical/low] (D) If one considered 10 000 similar cases, 0.7% would kill someone within a month if not admitted to hospital – a low risk’’. [Percentage/low]
21
and provided them with a short explanation of the study and the rationale behind the research programme. They were also provided with the contact details for the researcher. Conditions where risk was presented as a frequency per 10 000 were termed ‘‘numerical’’ conditions and those where risk was presented as a percentage (0.7%), as ‘‘percentage’’ conditions. In addition, semantic labels of ‘‘low’’ and ‘‘high’’ were also applied. This resulted in four conditions: numerical/ high, percentage/high, numerical/low and percentage/low (shown in their full form above). The rationale behind using the values of 70 and 0.7% (both equalling the same number when applied to a population of 10 000 cases) was to avoid floor or ceiling effects in the data. Setting the number too low may have caused most respondents to respond with a decision to admit and vice versa if the number was set too high. Pilot testing supported this. Data analysis Admission decisions were analysed using a Chi2 analysis on each comparison to determine any statistical differences in admission decisions between psychiatrists. Ethical approval In accordance with UK NHS ethical procedures, since the research was low-risk and involving NHS staff, it was exempt from research ethics committee approval. Research and Development approval was given by Hertfordshire Partnership University NHS Foundation Trust, and local approval was sought from the RCPsych to send the questionnaire to their members. Consent was implied by the voluntary completion of the questionnaire.
Results Responses of ‘‘recommend admission’’ and ‘‘not recommend admission’’ between conditions show an ordering of responses as follows A4B & C4D (see Figure 1). When conditions were collapsed into numerical vs. percentage conditions, individuals were more likely to recommend admission in the numerical than percentage condition (X2 (1, N ¼ 678) ¼ 7.43, p ¼ 0.006). Moreover, analysis
The statement ‘‘No further information is available beyond this extract’’ concluded the vignette text. Participants then viewed the following text: ‘‘Question: Based on this limited information would you:’’ –Recommend admission –Not recommend admission Participants were required to select one of these two options. The final screen thanked them for their participation
Figure 1. Distribution of ‘‘recommend admission’’ and ‘‘not recommend admission’’ scores across conditions.
22
K. Jefferies-Sewell et al.
based on semantic label results suggested a statistically higher rate of admission in conditions where the ‘‘high risk’’ term was applied (X2 (1, N ¼ 372) ¼ 7.269, p ¼ 0.007).
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
Demographic differences Male participants recommended admission more often than females (57.29% [n ¼ 228] vs. 51.43% [n ¼ 144]) across all conditions, a difference that reached statistical significance (X2 (1, N ¼ 678) ¼ 4.34, p ¼ 0.03), albeit with a small effect size. When comparing demographics within the group who, under the numerical/high condition, chose not to admit a patient it was found that the group contained more males (n ¼ 33) than females (n ¼ 23). Furthermore, the group consisted of a greater number of older and less experienced participants.
Discussion Psychiatrists were more likely to recommend admission of a patient when risk information was presented as a frequency rather than as a percentage. This held true irrespective of whether the semantic label ‘‘high’’ or ‘‘low’’ was applied. The two other conditions (numerical/low and percentage/ high) showed almost identical patterns, which may be explained by their potentially competing biases. Based on demographic information, results showed a higher instance of a ‘‘recommend admission’’ decision in male psychiatrists, a pattern that is incompatible with the oft accepted notion that females are more risk averse than males (Jianakoplos & Bernasek, 1998). Experience had an effect on decision making with younger, less experienced psychiatrists choosing to admit the patient more often than older, more experienced psychiatrists. This study, using a large sample, provides strong evidence that psychiatrists are influenced by the way risk information is presented. There are few studies in the mental health literature that have explored the impact of such framing effects on decision-making, and so this research provides a first step in raising awareness of its importance. On this basis, future research may explore the circumstances under which individuals do not succumb to framing effects and may help inform appropriate training, particularly for less experienced professionals. A caveat to these results is that clinicians would usually have a richer source of information available on which to base their decisions. Participants were informed that they were taking part in a study investigating clinical decision-making but were not informed that they were only participating in one of the four conditions. We cannot know for sure what other factors intervened in the decision-making process. Moreover, although we recruited a large sample, we did not employ systematic sampling and therefore sex comparisons within the group and generalisations to the wider population should be cautiously interpreted. Our study employed a similar manipulation of risk information to Yamagishi (1997), who used ‘‘risk of death’’ probabilities presented as frequencies or percentages over different conditions and asked participants to rate riskiness of
J Ment Health, 2015; 24(1): 20–23
death by various diseases. Risk was perceived as being higher when presented as relative frequencies (numbers) using a larger range (1286 of 10 000). Correspondingly, our study found that presenting risk as a frequency (rather than percentage) increased caution, as indicated by a response to admit the patient. Although the studies differed in some respects (we also applied semantic labels), a similar result was found with regard to base-rate neglect. In Yamagishi’s experiments, participants tended to disregard the sample size and focussed on the actual number of deaths. Finally, and importantly, our study participants were psychiatrists, whereas the Yamagishi study recruited a university student sample. In our study, participants were also subject to a further manipulation using semantic labelling. Psychiatrists were influenced by the semantic labels ‘‘high’’ and ‘‘low’’ with those viewing ‘‘high’’ risk scenarios recommending admission more often than did those who saw the risk described as ‘‘low’’. Compared with the classical framing effect (Tversky & Kahneman, 1981) where situations positively framed as ‘‘lives gained’’ (i.e. individuals will survive) resulted in risk aversion and negatively framed situations resulted in risk seeking, there appears to be a medical version of the framework yielding a reverse pattern. This was also reported in a study by McNeil et al. (1982) who investigated decisions about hypothetical treatments (radiation or surgery). They found participants were more likely to choose the more invasive surgery option when the situation was framed in terms of ‘‘survival’’ (this being the positive, low risk frame) and radiation when viewed in terms of mortality (negative, high risk). Similarly, in our study, participants who were told the case was ‘‘high’’ risk were more inclined to err on the side of caution, whereas those who were told it was ‘‘low’’ risk were more willing to risk the safety of the patient and those he came into contact with. The significant framing effects in this task are likely explainable by a lack of understanding about probability. Although psychiatrists are routinely faced with probabilistic information (e.g. side effect profiles, NNTs, odds ratios), they may lack a clear understanding of what this means. Previous studies highlight a misunderstanding of basic probability not only by the general public (Kirsch et al., 2002) but also within a range of MHPs (Gale et al., 2003). In the latter study, MHPs were no more likely than education-, sex- and age-matched controls to correctly solve probability-related problems. In this study, we extend this work into a clinical context. The resulting bias towards an admission decision when the term ‘‘high-risk’’ was used may be explained by ‘‘attentional bias’’, i.e. a cognitive bias resulting in attention being focussed on semantic information. This may derive from increased pressure on clinicians to make accurate decisions in situations that are inherently ambiguous and uncertain. If it is the case that semantic labels can influence decisions to recommend admission, the implications could be reduced bed availability, increased workload and financial burden for NHS trusts. It is notable that many risk assessment proformas in mental health do use semantic terms such as high-, mediumand low-risk (Hawley et al., 2006). In this study, we sought to determine whether psychiatrists may be influenced by small changes in the presentation of risk
DOI: 10.3109/09638237.2014.951477
information. Our findings offer some initial insight into the presence of framing effects in a mental health decision making context and raise new questions for future research. For example, what kind of decisions are affected by framing? What are psychiatrists’ attitudes toward working with risk based information? Further research into these areas would undoubtedly be a valuable contribution to the literature.
Declaration of interest
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
This study was completed as part of a research assistant/PhD post funded by Hertfordshire Partnership University NHS Foundation Trust.
References Bloomfield AN. (2006). Group size and the framing effect. Mem Cognit, 34, 929–37.
Effect of framing on admission decisions
23
Gale TM, Hawley CJ, Sivakumaran T. (2003). Do mental health professionals really understand probability? Implications for risk assessment and evidence-based practice. J Ment Health, 12, 411–24. Hawley CJ, Littlechild B, Sivakumaran T, et al. (2006). Structure and content of risk assessment proformas in mental healthcare. J Ment Health, 15, 437–48. Jianakoplos NA, Bernasek A. (1998). Are women more risk averse? Econ Inq, 36, 620–30. Kirsch IS, Jungeblut A, Jenkins L, Kolstad A. (2002). Adult literacy in America: A first look at the findings of the National Adult Literacy Survey (NCES Report No. 1993-275; 3rd ed.).Washington, DC: U.S. Department of Education, National Center for Education Statistics. Marteau TM. (1989). Framing of information: Its influence upon decisions of doctors and patients. Br J Soc Psychol, 28, 89–94. McNeil BJ, Pauker SG, Sox Jr HC, Tversky A. (1982). On the elicitation of preferences for alternative therapies. N Engl J Med, 306, 1259–62. Tversky A, Kahneman D. (1981). The framing of decisions and the psychology of choice. Science, 211, 453–8. Wang X, Simons F, Bre´dart S. (2001). Social cues and verbal framing in risky choice. J Behav Decis Making, 14, 1–15. Yamagishi K. (1997). When a 12.86% mortality is more dangerous than 24.14%: Implications for risk communication. Appl Cognit Psychol, 11, 495–506.
ISSN: 0963-8237 (Print) 1360-0567 (Online) Journal homepage: http://www.tandfonline.com/loi/ijmh20
To admit or not to admit? The effect of framing on risk assessment decision making in psychiatrists Kiri Jefferies-Sewell, Shivani Sharma, Tim M. Gale, Chris J. Hawley, George J. Georgiou & Keith R. Laws To cite this article: Kiri Jefferies-Sewell, Shivani Sharma, Tim M. Gale, Chris J. Hawley, George J. Georgiou & Keith R. Laws (2015) To admit or not to admit? The effect of framing on risk assessment decision making in psychiatrists, Journal of Mental Health, 24:1, 20-23 To link to this article: http://dx.doi.org/10.3109/09638237.2014.951477
Published online: 04 Sep 2014.
Submit your article to this journal
Article views: 134
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=ijmh20 Download by: [University of California, San Diego]
Date: 05 November 2015, At: 17:42
http://informahealthcare.com/jmh ISSN: 0963-8237 (print), 1360-0567 (electronic) J Ment Health, 2015; 24(1): 20–23 ! 2015 Shadowfax Publishing and Informa UK Limited. DOI: 10.3109/09638237.2014.951477
ORIGINAL ARTICLE
To admit or not to admit? The effect of framing on risk assessment decision making in psychiatrists Kiri Jefferies-Sewell1,2, Shivani Sharma2, Tim M. Gale1,2, Chris J. Hawley2, George J. Georgiou2, and Keith R. Laws2 1
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
Hertfordshire Partnership University NHS Foundation Trust, Mental Health Unit, QEII Hospital, Howlands, Welwyn Garden City, Hertfordshire, UK and 2Department of Psychology, School of Life and Medical Sciences, University of Hertfordshire, Hatfield, Hertfordshire, UK Abstract
Keywords
Background: The way that information is presented is well known to induce a range of biases in human decision tasks. Little research exists on framing effects in psychiatric decision making, but it is reasonable to assume that psychiatrists are not immune and, if so, there may be implications for the welfare of patients, staff and the general public. Aims: To investigate whether presentation of risk information in different formats (frequency, percentage and semantic) influences inpatient admission decisions by psychiatrists. Methods: Six-hundred seventy-eight general adult psychiatrists read a short clinical vignette presenting a case scenario of a patient presenting for inpatient admission. One of four condition questions followed the vignette, incorporating either numerical or percentage probabilities and the semantic labels ‘‘high’’ and ‘‘low’’ risk. In each condition, the actual risk was identical, but the way it was presented varied. The decision to admit the patient or not was recorded and compared across conditions. Results: More individuals chose to admit the patient when risk information was presented in numerical form (X2 ¼ 7.43, p ¼ 0.006) and with the semantic label ‘‘high’’ (X2 ¼ 7.27, p ¼ 0.007). Conclusions: Presentation of risk information may influence decision making in psychiatrists. This has important implications for mental health clinical practice where clinicians are required to interpret probabilistic information within their daily work.
Admissions, decision making, framing, psychiatry
Introduction Mental health professionals (MHPs) make important patient-related decisions on a daily basis, and so a clear understanding of their decision-making processes and biases is imperative to achieving best practice. Research has shown that, in general, human decision-making can be manipulated toward both risk aversion, and risk seeking, when information about risk factors is presented in different ways (Tversky & Kahneman, 1981), and this bias has been shown to extend to medical settings (Marteau, 1989; McNeil et al., 1982). This suggests that decisions made by health professionals are not independent of the information presentation format, a fact that may have implications for patient care. More recent findings suggest that framing effects appear only when the subject matter has real value for the decision maker (Bloomfield, 2006), and the direction of the effect is dependent on the baseline level of risk presented as well as the decision context (Wang et al., 2001). Yamagishi (1997) illustrated that manipulation of base-rate information and Correspondence: Kiri Jefferies-Sewell, Learning and Development Centre, Hertfordshire Partnership University NHS Foundation Trust, The Colonnades, Beaconsfield Close, Hatfield, AL10 8YD, UK. Tel: 01707 253 834. E-mail: [email protected]
History Received 26 September 2013 Revised 18 February 2014 Accepted 27 May 2014 Published online 4 September 2014
anchoring points influenced rankings of ‘‘risk of death’’. The author showed, in two experiments, that participants judged the probability 1286 out of 10 000 (12.86% chance) more risky than 24.14 in 100 (24.14% chance). The aim of this study was to investigate what influence the presentation of risk information has on inpatient admission decisions by psychiatrists. Specifically, we explored the manipulation of numerical (frequency per 10 000) vs. percentage data and the impact of using the semantic labels of ‘‘high’’ vs. ‘‘low’’ risk admission decisions. Based on previous work, we predict that presenting risk information numerically (as opposed to percentages) and using the term ‘‘high risk’’ will produce the highest number of admission decisions.
Methods Six-hundred seventy-eight (398 male) general adult psychiatrists recruited from the Royal College of Psychiatrists (RCPsych) and Hertfordshire Partnership University NHS Foundation Trust (HPFT) participated in this study. Psychiatrists from HPFT would also have had membership with RCPsych, but this information was not requested, and recruitment was sought from both the Trust and the RCP. However, comparison of anonymous collector identification
Effect of framing on admission decisions
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
DOI: 10.3109/09638237.2014.951477
numbers confirmed that none of the participants completed the questionnaire more than once. Using an independent measures design, all participants were randomly assigned to one of four conditions and e-mailed the relevant link. The presentation of risk data was manipulated to be presented as either a frequency per 10 000 or as a percentage. Semantic labels of ‘‘high’’ and ‘‘low’’ were also inserted to indicate an expert’s opinion about the level of risk. Prior to this experiment, the questionnaire was piloted on eight consultant and junior psychiatrists during a routine teaching session. The rationale behind this phase of the study was to ensure that decisions within conditions were not polarised. The ‘‘numerical high’’ condition (see procedure section for explanation of each condition) was presented to all participants, and results showed that 50% of participants chose to admit. Psychiatrists accessed an internet link to the online site where they were able to complete the questionnaire at their convenience. After entering demographic information (age, sex, profession, area of profession and length of time in role), participants then read the vignette (same for all participants) describing a patient who presents as follows: The patient is a 33-year-old male. At the age of 18 a firm diagnosis of schizophrenia, paranoid type, was made. The clinical course over the past 15 years has included readmission, some suicidal behaviour and some threatening behaviour to others. He presents to you with paranoia needing prompt treatment attention. Imagine you have the benefit of an extract from a report made by a previous treating Unit that had known the patient for a long time and has sophisticated, and highly accurate, techniques for assessing key risks. This text was followed by one of the following four conditions, to which participants were randomly assigned: (A) ‘‘If one considered 10 000 similar cases, 70 would kill someone within a month if not admitted to hospital – a high risk’’. [Numerical/high] (B) ‘‘If one considered 10 000 similar cases, 0.7% would kill someone within a month if not admitted to hospital – a high risk’’. [Percentage/high] (C) ‘‘If one considered 10 000 similar cases, 70 would kill someone within a month if not admitted to hospital – a low risk’’. [Numerical/low] (D) If one considered 10 000 similar cases, 0.7% would kill someone within a month if not admitted to hospital – a low risk’’. [Percentage/low]
21
and provided them with a short explanation of the study and the rationale behind the research programme. They were also provided with the contact details for the researcher. Conditions where risk was presented as a frequency per 10 000 were termed ‘‘numerical’’ conditions and those where risk was presented as a percentage (0.7%), as ‘‘percentage’’ conditions. In addition, semantic labels of ‘‘low’’ and ‘‘high’’ were also applied. This resulted in four conditions: numerical/ high, percentage/high, numerical/low and percentage/low (shown in their full form above). The rationale behind using the values of 70 and 0.7% (both equalling the same number when applied to a population of 10 000 cases) was to avoid floor or ceiling effects in the data. Setting the number too low may have caused most respondents to respond with a decision to admit and vice versa if the number was set too high. Pilot testing supported this. Data analysis Admission decisions were analysed using a Chi2 analysis on each comparison to determine any statistical differences in admission decisions between psychiatrists. Ethical approval In accordance with UK NHS ethical procedures, since the research was low-risk and involving NHS staff, it was exempt from research ethics committee approval. Research and Development approval was given by Hertfordshire Partnership University NHS Foundation Trust, and local approval was sought from the RCPsych to send the questionnaire to their members. Consent was implied by the voluntary completion of the questionnaire.
Results Responses of ‘‘recommend admission’’ and ‘‘not recommend admission’’ between conditions show an ordering of responses as follows A4B & C4D (see Figure 1). When conditions were collapsed into numerical vs. percentage conditions, individuals were more likely to recommend admission in the numerical than percentage condition (X2 (1, N ¼ 678) ¼ 7.43, p ¼ 0.006). Moreover, analysis
The statement ‘‘No further information is available beyond this extract’’ concluded the vignette text. Participants then viewed the following text: ‘‘Question: Based on this limited information would you:’’ –Recommend admission –Not recommend admission Participants were required to select one of these two options. The final screen thanked them for their participation
Figure 1. Distribution of ‘‘recommend admission’’ and ‘‘not recommend admission’’ scores across conditions.
22
K. Jefferies-Sewell et al.
based on semantic label results suggested a statistically higher rate of admission in conditions where the ‘‘high risk’’ term was applied (X2 (1, N ¼ 372) ¼ 7.269, p ¼ 0.007).
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
Demographic differences Male participants recommended admission more often than females (57.29% [n ¼ 228] vs. 51.43% [n ¼ 144]) across all conditions, a difference that reached statistical significance (X2 (1, N ¼ 678) ¼ 4.34, p ¼ 0.03), albeit with a small effect size. When comparing demographics within the group who, under the numerical/high condition, chose not to admit a patient it was found that the group contained more males (n ¼ 33) than females (n ¼ 23). Furthermore, the group consisted of a greater number of older and less experienced participants.
Discussion Psychiatrists were more likely to recommend admission of a patient when risk information was presented as a frequency rather than as a percentage. This held true irrespective of whether the semantic label ‘‘high’’ or ‘‘low’’ was applied. The two other conditions (numerical/low and percentage/ high) showed almost identical patterns, which may be explained by their potentially competing biases. Based on demographic information, results showed a higher instance of a ‘‘recommend admission’’ decision in male psychiatrists, a pattern that is incompatible with the oft accepted notion that females are more risk averse than males (Jianakoplos & Bernasek, 1998). Experience had an effect on decision making with younger, less experienced psychiatrists choosing to admit the patient more often than older, more experienced psychiatrists. This study, using a large sample, provides strong evidence that psychiatrists are influenced by the way risk information is presented. There are few studies in the mental health literature that have explored the impact of such framing effects on decision-making, and so this research provides a first step in raising awareness of its importance. On this basis, future research may explore the circumstances under which individuals do not succumb to framing effects and may help inform appropriate training, particularly for less experienced professionals. A caveat to these results is that clinicians would usually have a richer source of information available on which to base their decisions. Participants were informed that they were taking part in a study investigating clinical decision-making but were not informed that they were only participating in one of the four conditions. We cannot know for sure what other factors intervened in the decision-making process. Moreover, although we recruited a large sample, we did not employ systematic sampling and therefore sex comparisons within the group and generalisations to the wider population should be cautiously interpreted. Our study employed a similar manipulation of risk information to Yamagishi (1997), who used ‘‘risk of death’’ probabilities presented as frequencies or percentages over different conditions and asked participants to rate riskiness of
J Ment Health, 2015; 24(1): 20–23
death by various diseases. Risk was perceived as being higher when presented as relative frequencies (numbers) using a larger range (1286 of 10 000). Correspondingly, our study found that presenting risk as a frequency (rather than percentage) increased caution, as indicated by a response to admit the patient. Although the studies differed in some respects (we also applied semantic labels), a similar result was found with regard to base-rate neglect. In Yamagishi’s experiments, participants tended to disregard the sample size and focussed on the actual number of deaths. Finally, and importantly, our study participants were psychiatrists, whereas the Yamagishi study recruited a university student sample. In our study, participants were also subject to a further manipulation using semantic labelling. Psychiatrists were influenced by the semantic labels ‘‘high’’ and ‘‘low’’ with those viewing ‘‘high’’ risk scenarios recommending admission more often than did those who saw the risk described as ‘‘low’’. Compared with the classical framing effect (Tversky & Kahneman, 1981) where situations positively framed as ‘‘lives gained’’ (i.e. individuals will survive) resulted in risk aversion and negatively framed situations resulted in risk seeking, there appears to be a medical version of the framework yielding a reverse pattern. This was also reported in a study by McNeil et al. (1982) who investigated decisions about hypothetical treatments (radiation or surgery). They found participants were more likely to choose the more invasive surgery option when the situation was framed in terms of ‘‘survival’’ (this being the positive, low risk frame) and radiation when viewed in terms of mortality (negative, high risk). Similarly, in our study, participants who were told the case was ‘‘high’’ risk were more inclined to err on the side of caution, whereas those who were told it was ‘‘low’’ risk were more willing to risk the safety of the patient and those he came into contact with. The significant framing effects in this task are likely explainable by a lack of understanding about probability. Although psychiatrists are routinely faced with probabilistic information (e.g. side effect profiles, NNTs, odds ratios), they may lack a clear understanding of what this means. Previous studies highlight a misunderstanding of basic probability not only by the general public (Kirsch et al., 2002) but also within a range of MHPs (Gale et al., 2003). In the latter study, MHPs were no more likely than education-, sex- and age-matched controls to correctly solve probability-related problems. In this study, we extend this work into a clinical context. The resulting bias towards an admission decision when the term ‘‘high-risk’’ was used may be explained by ‘‘attentional bias’’, i.e. a cognitive bias resulting in attention being focussed on semantic information. This may derive from increased pressure on clinicians to make accurate decisions in situations that are inherently ambiguous and uncertain. If it is the case that semantic labels can influence decisions to recommend admission, the implications could be reduced bed availability, increased workload and financial burden for NHS trusts. It is notable that many risk assessment proformas in mental health do use semantic terms such as high-, mediumand low-risk (Hawley et al., 2006). In this study, we sought to determine whether psychiatrists may be influenced by small changes in the presentation of risk
DOI: 10.3109/09638237.2014.951477
information. Our findings offer some initial insight into the presence of framing effects in a mental health decision making context and raise new questions for future research. For example, what kind of decisions are affected by framing? What are psychiatrists’ attitudes toward working with risk based information? Further research into these areas would undoubtedly be a valuable contribution to the literature.
Declaration of interest
Downloaded by [University of California, San Diego] at 17:42 05 November 2015
This study was completed as part of a research assistant/PhD post funded by Hertfordshire Partnership University NHS Foundation Trust.
References Bloomfield AN. (2006). Group size and the framing effect. Mem Cognit, 34, 929–37.
Effect of framing on admission decisions
23
Gale TM, Hawley CJ, Sivakumaran T. (2003). Do mental health professionals really understand probability? Implications for risk assessment and evidence-based practice. J Ment Health, 12, 411–24. Hawley CJ, Littlechild B, Sivakumaran T, et al. (2006). Structure and content of risk assessment proformas in mental healthcare. J Ment Health, 15, 437–48. Jianakoplos NA, Bernasek A. (1998). Are women more risk averse? Econ Inq, 36, 620–30. Kirsch IS, Jungeblut A, Jenkins L, Kolstad A. (2002). Adult literacy in America: A first look at the findings of the National Adult Literacy Survey (NCES Report No. 1993-275; 3rd ed.).Washington, DC: U.S. Department of Education, National Center for Education Statistics. Marteau TM. (1989). Framing of information: Its influence upon decisions of doctors and patients. Br J Soc Psychol, 28, 89–94. McNeil BJ, Pauker SG, Sox Jr HC, Tversky A. (1982). On the elicitation of preferences for alternative therapies. N Engl J Med, 306, 1259–62. Tversky A, Kahneman D. (1981). The framing of decisions and the psychology of choice. Science, 211, 453–8. Wang X, Simons F, Bre´dart S. (2001). Social cues and verbal framing in risky choice. J Behav Decis Making, 14, 1–15. Yamagishi K. (1997). When a 12.86% mortality is more dangerous than 24.14%: Implications for risk communication. Appl Cognit Psychol, 11, 495–506.
