J CUII Epi&miol Vol. 44, No. 8, pp. 771-777, Printed in Great Britain
0895-4356/91 $3.00 + 0.00 Pergamon Press plc
1991
A RANDOMIZED QUANTITATIVE DAVID
CONTROLLED TRIAL COMPARING INFORMED CONSENT FORMATS* L. SWELL and JOHNR. FEUSSNER~
Ambulatory Care Service and Center for Health Services Research in Primary Care, Durham Veterans Administration Medical Center, and Division of General Internal Medicine, Duke University Medical Center, Durham, North Carolina, U.S.A. (Received in revised form 19 December 1990)
Abstract-Informed consent has been indirectly studied only in settings that do not replicate the actual consent process. We designed a sham study and randomly allocated adult ambulatory patients to receive one of two consent forms: Consent A (n = 52) described a randomized trial of usual treatment vs a new medication that “may work twice as fast as the usual treatment”; or Consent B (n = 48) that described a randomized trial of a new medication that “may work half as fast as the usual treatment”. Patients randomized to Consent A were more likely to consent than those randomized to Consent B (consent rate A = 67%, consent rate B = 42%, p < 0.01). Among patients who cited quantitative information, the difference in consent rate was even more marked (95% vs 36%, p c 0.001); patients who did not cite quantitative information had equivalent consent rates. Patients who perceived minimal or severe symptoms had lower consent rates than those with mid-range symptom scores (xi = 8.35, p = 0.015). Patients who recognize quantitative information will use it to make informed consent decisions. Informed consent
Randomized controlled trial
INTRODUCTION
Informed consent is a maxim of clinical care and research involving disclosure, comprehension, voluntariness, competence, and finally consent [l-3]. Empirical studies of informed consent may lead to a better understanding of the consent process by describing patient characteristics and factors that affect consent rates. The current process of informed consent places a major responsibility on the patient-they must process the information in a consent form and they must be able to think probabilistically [3].
*Presented at the Swtbani See&r of the Sodety for Cenerpl
InternalMedidae, New Orleans, Louisiana, 1 February 1989. tCurrent address: Center for Health Services Research in Primary Care, Durham Veterans Administration Medical Center, Durham, NC 27705, U.S.A. Address
correspondence to David L. Simel.
Decision making
Despite the requirements for informed consent during clinical care and research, there have b een few empiric trials of the actual decisionmaking process during consent. Assessing patients’ competence and judgment presents the most frequent and complex clinical problem of informed consent. Adequate methods for studying these complicated aspects of informed consent do not exist. While the d eve 1opment of instruments to assess competence is an active research area [4-6], the lack of such instruments limits research in actual clinical situations. Limitations of previous empiric informed consent research included use of nonpatient subjects or hypothetical situations, failure to consider what physicians actually told the patient, focusing only on patient recall as the key to understanding, lack of generalizability, or allowing investigator bias where the investigator was also the subject’s physician [l].
771
772
DAVIDL. SIMELand JOHN R. FEUSSNER
Accordingly, we created a consent form for a fictitious medication and used it to empirically evaluate the informed consent process in an ambulatory care clinic. The consent forms specified either that the new medication “may work twice as fast” or “may work half as fast” as the usual treatment. Although evaluating probabilistic information may sometimes be difficult [7,8], the ability to process the simple information in the consent form phrases “twice as fast” or “half as fast” was central to making an informed decision. Our study was designed to minimize most of the limitations of existing empirical trials of informed consent. Our specific aims were to: (1) assess the impact on subsequent consent rates of simple quantitative information in a consent form; (2) determine whether citation of quantitative information impacts on subsequent consent rates; (3) evaluate any interaction between the consent form and citation of quantitative information; and (4) determine the impact of symptom severity on consent rate. The hypotheses were: (I) the consent rates would be equal when patients did not cite quantitative information in the consent form as a factor in their decision; and (II) the consent rates would be different when patients did cite quantitative information as a factor in their decision to participate. METHODS
Study site The investigation was conducted in the Durham Veterans Administration Medical Center, a Deans’ Committee affiliate hospital of Duke University Medical Center. Patients were recruited from a walk-in ambulatory care clinic where 50-60 patients are evaluated daily for a variety of non-emergent medical and minor surgical problems. All patients were examined by a postgraduate year 2 or 3 Duke University Medical Houseofficer, or a general internist attending physician. Patient selection Subjects were identified after completing their clinical evaluation. Patients were included if they did not require hospital admission and were prescribed a new medication for treatment of at least one of their presenting problems. Patients were excluded if they had clinically apparent psychosis, delirium, a previous diag nosis of dementia, or had already participated in the study. No patients knew the physician-
investigator before agreeing to participate in the study. Information collected at baseline included sociodemographic variables, reason for visit data, and symptom severity. Symptom severity was assessed in the final 59 subjects using a 10 cm visual analog scale (0 = “barely noticeable”, 10 = “worst symptoms ever”). Study design Consecutive patients fulfllling the inclusion criteria on designated study days were asked to participate and offered the opportunity to discuss a new medication, for their presenting problem, with the investigator. No patient refused to participate. For example, a patient prescribed an oral analgesic for headaches would be offered the opportunity to participate in a trial evaluating a new oral analgesic for headaches, while a patient with pruritus prescribed topical agents would be offered participation in a trial evaluating a new topical treatment for pruritus. Patients prescribed more than one medication were asked to participate in a trial for a new medication that would replace only one of their current medications. Patients were not told that the “new” medication was fictitious. Patients were randomly assigned to Consent Form A or B (see the Appendix) using a computer-generated scheme. Consents A and B were identical except that Consent A stated “the new medication, AlOO, may work twice as fast . . . “, whereas Consent B stated “the new medication, AlOO, may work half as fast. . . “. After obtaining baseline information each patient was asked if they would permit a confidential tape-recording of the interview for the purpose of record keeping; 98% agreed to allow taping. The consent form was read aloud to subjects by the same investigator as the patients followed the text. Each reading of the consent form was preceded by the statement “I want you to understand that it is most important to me that you do what is right for you. Do not worry about pleasing me, the Veterans Administration, or anyone else. You do what is right for you,” Patients were given the opportunity to voluntarily ask questions or discuss the trial before deciding whether or not to participate. If the patient did not ask questions or state a consent preference after the consent was read they were prompted with the question, “How do you feel about this study?” Each patient was also asked what particular information helped them make their decision about participation.
713
Randomized Controlled Trial of Informed Consent
Patients who voluntarily mentioned the medication’s speed of action (“twice as fast” or “half as fast”) were considered by the investigator to have cited quantitative information as a factor in their decision whether or not to participate, even if speed of action was not the reason for their decision. The investigator reviewed briefly the chart of all patients who consented to participate and then informed them that their medical record suggested it would be best if they received the already prescribed “usual care”. Patients were never told that the consent forms were for non-existent studies. The study was approved by the Research and Development Committee and the Human Studies Committee of the Durham Veterans Administration Medical Center. The Research Committee reviews all proposals for scientific content, whereas the Human Studies Committee reviews ethical implications of proposed research. The Human Studies Committee is a multidisciplinary group composed of scientists, theologians, and lay persons from the community. The review committees conducted extensive discussion regarding the decision not to debrief the patients since deception studies, whenever possible, should allow subjects to be informed of the deception at the conclusion of the study. Deception research is ethically and legally permissible under certain circumstances: there is no more than minimal risk; the subjects’ rights and welfare are not affected adversely; and the research can not be otherwise performed [2]. The investigators preferred to conduct the study without debriefing, but asked the Human Studies Committee for guidance. Our preference arose from one of the perceived strengths of our study-the lack of investigator bias. By design, the patients in our study would not know the investigator, but we did not know how familiar they would be with our hospital. Therefore, we did not want to risk the patients’ inferring that when they came to the hospital for care, their doctor’s might not be trustworthy. Our concern that engendering mistrust might impact on future health care led us to conclude that immediate debriefing would be more risky than maintaining the deception. In view of the minimal potential for adverse consequences from the deception, the Human Studies Committee concluded that debriefing the patients was not necessary. Independent
review
A practicing attorney, who was blinded to the study hypotheses, reviewed a 25% (n = 25)
random sample of physician-patient encounters. Tape recordings were reviewed to obtain the attorney’s opinion concerning the fairness of the physician’s overall presentations. Specifically, the attorney assessed whether patients were unduly influenced to either consent or decline participation. In all cases, the independent attorney rated presentations as fair and balanced. The attorney was also asked to judge whether quantitative information was a factor in the patient’s decision to participate. The attorney’s judgment concerning citation of quantitative information agreed with the investigators in 88% of cases (IC= 0.76). Data analysis
Continuous variables were described as medians and analyzed using non-parametric tests (Wilcoxon rank sum) to determine if variables were distributed equally between groups receiving Consent A or B (a = 0.05). Dichotomous variables were analyzed with X2-tests to determine if the variables were distributed equally within Consent A or B, and between those that consented or declined. Logistic models were created to determine whether any variables predicted patient consent; the contribution of variables to the model were assessed using likelihood ratio and Wald x2 statistics. A model was developed first by evaluating data related to the primary hypothesis and collected on all patients. After determining the model for the primary variables of interest, the contribution of the symptom score to the model was determined for a subset of patients (n = 59). Somer’s DYx, an index of rank correlation, was used to assess the correlation between predicted probability of consent and observed probability of consent for this final model [9]. This rank correlation, which corrects for ties on the predicted probability, has a value of 1 for perfect correlation and 0 for random correlation. All analyses were performed using SAS [9]. RESULTS
One hundred patients were enrolled in the study; 52 were randomized to Consent A and 48 to Consent B. Patients randomized to Consent A vs B were similar in age, years of education, race, employment status, and presence of a family member during the consent process (Table 1). Patients had a variety of problems that were distributed similarly within Consents A and B (Table 1). While patients randomized
774
D~vto L. S~MELand JOHNR. FEU~SNEX
Table 1. Presenting features of enrolled patients by consent form Patient Characteristic l Age (yr, median) Education (yr, median) Race (% black) Employment status (% employed) Family member present (%) Presenting problem Paint Pruritis/skin condition Dyspnea/cough Hypertension Prostatitis/hematuria Gastrointestinal$ Ear/nose/throat problem Miscellaneous#
Consent A (n = 52)
Consent B (n = 48)
58 12 58
51 10 42
33 2
31 6
29 8 3 1 1 4 4 2
28 4 4 5 4 1 0 2
*p-Values from Wilcoxon rank sum, x2 or Fisher exact all 30.1. TIncludes all disorders for which primary problem is pain (e.g. headache, chest, abdominal, or extremity pain). tconstipation, hematochexia, nausea. §Hypercholesterolemia, diabetes, viral syndrome.
to Consent A described more severe symptoms, median symptom scores were not significantly different after adjusting for multiple comparisons (A = 7.2 mm, B = 3.8 mm, p = 0.03; p < 0.01 required for statistical significance given 5 comparisons). The baseline demographic data remained comparable after patients were partitioned according to their consent preference (n = 55 consenting, n = 45 declining). We first analyzed outcomes dealing with the primary study aims concerning the consent form and citation of quantitative information. Patients randomized to consent A were more likely to consent than those randomized to Consent B (consent rate A = 67%, consent rate B = 42%, p < 0.01; Table 2A). However, patients who cited quantitative information were no more likely to consent to participation than those who did not cite quantitative information (cited quantitative information = 62% consent rate, did not cite quantitative information = 49% consent rate, p = 0.2; Table 2B). Next, we analyzed the raw data to determine if the main effects, type of consent form, and citation of quantitative information were associated with the dependent variable, consent rate (Table 3). Patients randomized to Consent A were more likely to consent to participate than those randomized to Consent B after controlling for citation of quantitative information (p < 0.01). However, the consent rate was strongly affected
Table 2. Impact of consent form and citation of quantitative information on subseouent consent decisions Consent Decision Consent Decline
(A)
Consent Form
55
45
Consent Decision Decline Consent
(B) Cited Quantitative Information
55
45
by an interaction between the consent form and whether the patient cited quantitative information when making a decision (interaction term fi = 3.43 for consent form = A and citation of quantitative information, x2 = 7.72, p < 0.005). Patients who cited quantitative information in their decision were highly likely to consent when randomized to Consent A (95% consented to participate) and unlikely to consent when randomized to Consent B [36% consented to participate (risk ratio = 2.6, p < O.OOl), Table 31. All the subjects who consented to participate after being offered Consent A indicated they wanted the faster medication, while subjects who declined Consent B did so usually (88%) because they did not want a medication that was only half as fast as the usual treatment (Table 4). Patients who did not cite quantitative information in their decision were equally likely to participate when randomized to Consent A (50% consented to participate) or Consent B Table 3. Interaction between consent form and citation of quantitative information on consent rate Cited Quantitative Information No Yes
Form
Consent Form Form
Randomized Controlled Trial of Informed Consent Table 4. Reason for decision given by patients who did cite quantitative information Frequency Cement A (“Twice as fast”) Patients who consented (n = 19)
19
Wanted a medicine that was twice as fast Patients who declined (n = I)
1
Wanted his usual medication Consent B (“Half as fast”) Patients who consented (n = 9)
2 1 1 1
Willing to try something new Slower is better in the long run Wanted the “challenge” Would do what the doctor wanted Doctor would not suggest anything bad Wanted all medications from the VA Willing to help the VA No particular reason
1 1 1 1
Pafients who declined (n = 16)
Wanted a faster medication Not enough known about the new medication Does not like new things
14 1 1
(48% consented to participate, risk ratio = 1.1, Table 3). The most frequent reason these patients consented to participate was their preference for newer treatments, while the most frequent reason for declining was fear of experimentation (Table 5). We were unable to detect differences in reasons because of the large number of different answers among relatively few patients. Logistic models were created to determine which variables were useful in predicting the consent rate other than randomized consent form, citation of quantitative information for decision making, and interaction between consent form and quantitative information. The patient’s age (linear and quadratic terms), years Table 5. Reason for decision given by patients who did not cite quantitative information Frequency Patients who consented (n = 27)
Prefer new “things” Willing to try anything for relief Thought it would help others No particular reasons cited Willing to try if the new medication worked Did not believe it would hurt Usual medication would not work Believed side effects less with new medication Patients
who declined (n = 28)
Did not like the idea of an experiment Usual medication seemed safer or better Symptoms not severe enough Wanted more information Prior bad experience in a drug trial Did not want any medication Did not want to wait to receive new medication Patient blind and could not see consent form Trusted physician who prescribed usual treatment No particular reason cited CEu,sn
9 8 3 2 2 1 1 1 14 5 2
1 1
1 1 1 1 1
175
of education, race, and whether additional questions were asked after presentation of the consent form added no further predictive information to the logistic model (x: = 3.18, p = 0.7). After determining the lack of effect for all these variables in the entire study population, the contribution of severity of symptoms was determined for the subset of 59 patients who used the visual analog scale to describe their symptoms. Severity of symptoms did not make a linear contribution to the model (x: = 0.16, p = 0.7). However, adding the linear and quadratic terms for analog-scaled severity of symptoms to the three-variable model (consent form, citation of quantitative information, and interaction term) demonstrated a significant effect (1: = 8.35, p = 0.015). The variables consent form, citation of quantitative information, their interaction, linear and quadratic analog score were jointly significantly associated with the consent rate (xi = 13.77, p = 0.017) and had good predictive ability for consent rate [Somer DYX= 0.5, Fig. l(A,B)]. The graphic relationship of the terms in the model demonstrates that subjects with minimal or severe symptoms were less likely to consent than those with “mid-range” symptoms. DISCUSSION
The results from our study suggested strongly that what the consent forms contained and what the patient assimilated from the consent process affected the decision to participate in a research trial. Study patients who cited the information that the experimental drug worked “twice as fast” were much more likely to consent (95% consented) than patients who cited the information that the experimental drug worked “half as fast” as usual treatment (36% consented). Patients who did not cite quantitative information had similar consent rates for Consent A vs B (consent rate A = 50%, B = 48%, relative risk for consent = 1.1). The impact of such simple, quantitative information on consent rates was impressive. We infer that adding other similarly simple quantitative information into current consent forms may be useful. Unfortunately, only 45% of our patients cited the quantitative information. In order for patients to make informed decisions, future research should evaluate better ways of enabling them to recognize and use quantitative information. The consent rates were modified significantly by the severity of symptoms measured on a
776
DAVIDL. SIMIZL and JOHNR. FE~NER
(A) Probability of Conrent
0’ 0
1
2
3
4
6
Analog
6
7
8
9
10
7
8
9
10
Score
(B) Probability of Conrent 1
0.8
0.8
0
1
2
3
4
6
8
Analog Score Fig. 1. The predicted probability of consent vs the analog score, grouped by whether or not quantitative infortuation was cited iu the decision: (A) patients who cited quantitative information and (B) patients who did not cite quantitative information; ~-Consent A, A--Consent B). Probability of consent=l/{l+exp[-(-1.19+0.05*F-0.72sQ+ 2.53 * F* Q + 0.9 + analog - 0.09 + analog*)]}. F: 1 = Consent A, 0 = Consent B. Q: 1 = cited quantitative information, 0 = Did not cite quantitative information.
visual analog scale [Fig. l(A,B)]. This relationship was non-linear: patients with minimal or intense symptoms were less likely to consent than those with moderately intense symptoms. Patients who perceived their symptoms as minimal or severe may have been “risk averse”. Patients with minimal symptoms may have believed the usual treatment was sufficient so that taking risks associated with an unknown medication were not justifiable. Subjects with severe symptoms may have believed their symptoms were so severe they would rather accept the
known usual treatment than risk the new medication, especially given the new medication’s uncertain efficacy. Inferences from this finding relate importantly to patients’ tolerance for risk, but should be validated in other groups of patients and other clinical settings. The lack of a linear trend between perceived symptom severity and consent rate supports strongly the process of informed consent by contradicting the view that severely ill patients feel coerced to try anything. We created a true experiment so that patients actually believed they were making a decision that might lead to participation in an actual trial. This study demonstrates that empiric trials of the informed consent process are feasible and can be conducted without many of the methodological limitations of previous work. Our evaluation of the informed consent process addressed the five important limitations of previous work outlined by Meisel and Roth [l]. First, we tape-recorded patient-physician interactions and submitted them for review by an attorney who was blinded to the study hypotheses. This allowed us to assess what the patient was actually told and to document that the presentation was unbiased. Second, we did not succumb to the conceptual error of only evaluating patient recall. Third, rather than employing non-patients in hypothetical situations, we used real patients who were faced with making a real medical decision. Fourth, our study should be generalizable to other non-emergent ambulatory settings by virtue of the diversity of presenting problems. Despite patient’s inferring they were participating in a trial, our study’s generalizability may be limited by our use of a clinically unlikely scenario. Finally, we avoided investigator bias pertaining to informed consent research by including only subjects who were not patients of the investigator. For methodological reasons, we enhanced the validity of our study by eliminating investigator bias. However, eliminating this bias adversely influenced our overall consent rate which was relatively low (55%). We believe that the patients’ unfamiliarity with the investigator, or their more skeptical view of the consent form contributed to our low consent rate. Because we eliminated investigator bias, we could not directly assess the influence the physician might have on obtaining consent from their own patients. We believe that physicians’ influence on the informed consent process is dramatic. Physicians can affect favorably patients’ consent
111
Randomized Controlled Trial of Informed Consent
to procedures, surgery, and clinical trials by acquiring their consent, or they can affect consent unfavorably by exerting too much influence on patients’ decisions. While our study demonstrated a significant effect on consent rate when quantitative information was cited, we also observed a broad gap between disclosure and decision-only 45% of patients cited quantitative information in their decision. Additional empirical studies should focus on ways to improve patient recognition of important components of consent and assimilation of these components into their decision making. In addition, the potentially powerful role played by the physician should be systematically evaluated in varying patient situations. The gap between disclosure and decision should not create barriers to clinical care or research; on the contrary, better understanding may serve to reduce barriers and ultimately improve consent rates for clinical research. Acknowledgement-This work was supported in part by a grant from the A. W. Mellon Foundation. APPENDIX INFORMED
REFERENCES
l. Meisel A, Roth LH. What we do and do not know
about informed consent. JAMA 1981; 246: 2473-2477. 2. Appelbaum PS, Lidz CW, Me&l A. Informed Cotmeotz Legal Theory and Clinical Practice. New York: Oxford 3 University Press; 1987. President’s Commission for the Study of Ethical ’ Problems in Medicine and Biomedical and Behavioral Research. Making Health Care De&ioos, Vol. 1. Washington, D.C.: U.S. Government Printing Office; 1982. 4. Kaplan KH. Assessing judgment. Gen Hasp Psycho1 1988; 9: 202-208. 5. Appelbaum PS, Grisso T. Assessing patient’s capacities to consent to treatment. N m J Med 1988: jl9: 1635-1638. 6. Grisso T. Evaluating competencies: forensic assessments and instruments. In: Sales BD. Ed. Perrmeetlvea la Law and Psychology, Vol. 7. New York: Plenum Press; 1986. 7. Tversky A, Kahneman D. The framing of decisions and the psychology of choice. Science 1981; 211: 453458. 8. Mazur DJ. Why the goals of informed consent are not realized: treatise on informed consent for the primary care physician. J Gen Intern Med 1988; 3: 370-380. 9. Harrell FE. The LOGIST procedure. In: SUGI Supplemental Library User’s Guide, Version 5. Cary, N.C.: SAS Institute Inc.; 1986: 269-293. A
CONSENT TO EVALUATE MEDICATION
Al00 IN RELIEVING
SYMPTOMS
Your doctors are currently evaluating you for . We are investigating the use of a new medication, AlOO, versus the usual treatment for patients with your condition. The use of either the new medication or the usual treatment has no known permanent side effects. If you agree to participate in this investigation, we will determine your ability to take the new medication from the examination and tests already performed. If the medication can be given to you safely we will decide whether you will receive the experimental medication or the usual treatment by chance. This is a common way of assessing new medications. The cost to you will be the same, whether or not you participate in the investigation. It is important that you understand the following: (1)
If you choose not to participate, your physician will provide you with the best available usual treatment.
(FOR CONSENT A): (2)
The new medication, AIOO, may work twice as fast as the usual treatment.
(FOR CONSENT B): (2) I AGREE/DO
Date
The new medication, AlOO, may work half as fast as the usual treatment. NOT AGREE (circle one or the other) to participate in this investigation:
Time
0895-4356/91 $3.00 + 0.00 Pergamon Press plc
1991
A RANDOMIZED QUANTITATIVE DAVID
CONTROLLED TRIAL COMPARING INFORMED CONSENT FORMATS* L. SWELL and JOHNR. FEUSSNER~
Ambulatory Care Service and Center for Health Services Research in Primary Care, Durham Veterans Administration Medical Center, and Division of General Internal Medicine, Duke University Medical Center, Durham, North Carolina, U.S.A. (Received in revised form 19 December 1990)
Abstract-Informed consent has been indirectly studied only in settings that do not replicate the actual consent process. We designed a sham study and randomly allocated adult ambulatory patients to receive one of two consent forms: Consent A (n = 52) described a randomized trial of usual treatment vs a new medication that “may work twice as fast as the usual treatment”; or Consent B (n = 48) that described a randomized trial of a new medication that “may work half as fast as the usual treatment”. Patients randomized to Consent A were more likely to consent than those randomized to Consent B (consent rate A = 67%, consent rate B = 42%, p < 0.01). Among patients who cited quantitative information, the difference in consent rate was even more marked (95% vs 36%, p c 0.001); patients who did not cite quantitative information had equivalent consent rates. Patients who perceived minimal or severe symptoms had lower consent rates than those with mid-range symptom scores (xi = 8.35, p = 0.015). Patients who recognize quantitative information will use it to make informed consent decisions. Informed consent
Randomized controlled trial
INTRODUCTION
Informed consent is a maxim of clinical care and research involving disclosure, comprehension, voluntariness, competence, and finally consent [l-3]. Empirical studies of informed consent may lead to a better understanding of the consent process by describing patient characteristics and factors that affect consent rates. The current process of informed consent places a major responsibility on the patient-they must process the information in a consent form and they must be able to think probabilistically [3].
*Presented at the Swtbani See&r of the Sodety for Cenerpl
InternalMedidae, New Orleans, Louisiana, 1 February 1989. tCurrent address: Center for Health Services Research in Primary Care, Durham Veterans Administration Medical Center, Durham, NC 27705, U.S.A. Address
correspondence to David L. Simel.
Decision making
Despite the requirements for informed consent during clinical care and research, there have b een few empiric trials of the actual decisionmaking process during consent. Assessing patients’ competence and judgment presents the most frequent and complex clinical problem of informed consent. Adequate methods for studying these complicated aspects of informed consent do not exist. While the d eve 1opment of instruments to assess competence is an active research area [4-6], the lack of such instruments limits research in actual clinical situations. Limitations of previous empiric informed consent research included use of nonpatient subjects or hypothetical situations, failure to consider what physicians actually told the patient, focusing only on patient recall as the key to understanding, lack of generalizability, or allowing investigator bias where the investigator was also the subject’s physician [l].
771
772
DAVIDL. SIMELand JOHN R. FEUSSNER
Accordingly, we created a consent form for a fictitious medication and used it to empirically evaluate the informed consent process in an ambulatory care clinic. The consent forms specified either that the new medication “may work twice as fast” or “may work half as fast” as the usual treatment. Although evaluating probabilistic information may sometimes be difficult [7,8], the ability to process the simple information in the consent form phrases “twice as fast” or “half as fast” was central to making an informed decision. Our study was designed to minimize most of the limitations of existing empirical trials of informed consent. Our specific aims were to: (1) assess the impact on subsequent consent rates of simple quantitative information in a consent form; (2) determine whether citation of quantitative information impacts on subsequent consent rates; (3) evaluate any interaction between the consent form and citation of quantitative information; and (4) determine the impact of symptom severity on consent rate. The hypotheses were: (I) the consent rates would be equal when patients did not cite quantitative information in the consent form as a factor in their decision; and (II) the consent rates would be different when patients did cite quantitative information as a factor in their decision to participate. METHODS
Study site The investigation was conducted in the Durham Veterans Administration Medical Center, a Deans’ Committee affiliate hospital of Duke University Medical Center. Patients were recruited from a walk-in ambulatory care clinic where 50-60 patients are evaluated daily for a variety of non-emergent medical and minor surgical problems. All patients were examined by a postgraduate year 2 or 3 Duke University Medical Houseofficer, or a general internist attending physician. Patient selection Subjects were identified after completing their clinical evaluation. Patients were included if they did not require hospital admission and were prescribed a new medication for treatment of at least one of their presenting problems. Patients were excluded if they had clinically apparent psychosis, delirium, a previous diag nosis of dementia, or had already participated in the study. No patients knew the physician-
investigator before agreeing to participate in the study. Information collected at baseline included sociodemographic variables, reason for visit data, and symptom severity. Symptom severity was assessed in the final 59 subjects using a 10 cm visual analog scale (0 = “barely noticeable”, 10 = “worst symptoms ever”). Study design Consecutive patients fulfllling the inclusion criteria on designated study days were asked to participate and offered the opportunity to discuss a new medication, for their presenting problem, with the investigator. No patient refused to participate. For example, a patient prescribed an oral analgesic for headaches would be offered the opportunity to participate in a trial evaluating a new oral analgesic for headaches, while a patient with pruritus prescribed topical agents would be offered participation in a trial evaluating a new topical treatment for pruritus. Patients prescribed more than one medication were asked to participate in a trial for a new medication that would replace only one of their current medications. Patients were not told that the “new” medication was fictitious. Patients were randomly assigned to Consent Form A or B (see the Appendix) using a computer-generated scheme. Consents A and B were identical except that Consent A stated “the new medication, AlOO, may work twice as fast . . . “, whereas Consent B stated “the new medication, AlOO, may work half as fast. . . “. After obtaining baseline information each patient was asked if they would permit a confidential tape-recording of the interview for the purpose of record keeping; 98% agreed to allow taping. The consent form was read aloud to subjects by the same investigator as the patients followed the text. Each reading of the consent form was preceded by the statement “I want you to understand that it is most important to me that you do what is right for you. Do not worry about pleasing me, the Veterans Administration, or anyone else. You do what is right for you,” Patients were given the opportunity to voluntarily ask questions or discuss the trial before deciding whether or not to participate. If the patient did not ask questions or state a consent preference after the consent was read they were prompted with the question, “How do you feel about this study?” Each patient was also asked what particular information helped them make their decision about participation.
713
Randomized Controlled Trial of Informed Consent
Patients who voluntarily mentioned the medication’s speed of action (“twice as fast” or “half as fast”) were considered by the investigator to have cited quantitative information as a factor in their decision whether or not to participate, even if speed of action was not the reason for their decision. The investigator reviewed briefly the chart of all patients who consented to participate and then informed them that their medical record suggested it would be best if they received the already prescribed “usual care”. Patients were never told that the consent forms were for non-existent studies. The study was approved by the Research and Development Committee and the Human Studies Committee of the Durham Veterans Administration Medical Center. The Research Committee reviews all proposals for scientific content, whereas the Human Studies Committee reviews ethical implications of proposed research. The Human Studies Committee is a multidisciplinary group composed of scientists, theologians, and lay persons from the community. The review committees conducted extensive discussion regarding the decision not to debrief the patients since deception studies, whenever possible, should allow subjects to be informed of the deception at the conclusion of the study. Deception research is ethically and legally permissible under certain circumstances: there is no more than minimal risk; the subjects’ rights and welfare are not affected adversely; and the research can not be otherwise performed [2]. The investigators preferred to conduct the study without debriefing, but asked the Human Studies Committee for guidance. Our preference arose from one of the perceived strengths of our study-the lack of investigator bias. By design, the patients in our study would not know the investigator, but we did not know how familiar they would be with our hospital. Therefore, we did not want to risk the patients’ inferring that when they came to the hospital for care, their doctor’s might not be trustworthy. Our concern that engendering mistrust might impact on future health care led us to conclude that immediate debriefing would be more risky than maintaining the deception. In view of the minimal potential for adverse consequences from the deception, the Human Studies Committee concluded that debriefing the patients was not necessary. Independent
review
A practicing attorney, who was blinded to the study hypotheses, reviewed a 25% (n = 25)
random sample of physician-patient encounters. Tape recordings were reviewed to obtain the attorney’s opinion concerning the fairness of the physician’s overall presentations. Specifically, the attorney assessed whether patients were unduly influenced to either consent or decline participation. In all cases, the independent attorney rated presentations as fair and balanced. The attorney was also asked to judge whether quantitative information was a factor in the patient’s decision to participate. The attorney’s judgment concerning citation of quantitative information agreed with the investigators in 88% of cases (IC= 0.76). Data analysis
Continuous variables were described as medians and analyzed using non-parametric tests (Wilcoxon rank sum) to determine if variables were distributed equally between groups receiving Consent A or B (a = 0.05). Dichotomous variables were analyzed with X2-tests to determine if the variables were distributed equally within Consent A or B, and between those that consented or declined. Logistic models were created to determine whether any variables predicted patient consent; the contribution of variables to the model were assessed using likelihood ratio and Wald x2 statistics. A model was developed first by evaluating data related to the primary hypothesis and collected on all patients. After determining the model for the primary variables of interest, the contribution of the symptom score to the model was determined for a subset of patients (n = 59). Somer’s DYx, an index of rank correlation, was used to assess the correlation between predicted probability of consent and observed probability of consent for this final model [9]. This rank correlation, which corrects for ties on the predicted probability, has a value of 1 for perfect correlation and 0 for random correlation. All analyses were performed using SAS [9]. RESULTS
One hundred patients were enrolled in the study; 52 were randomized to Consent A and 48 to Consent B. Patients randomized to Consent A vs B were similar in age, years of education, race, employment status, and presence of a family member during the consent process (Table 1). Patients had a variety of problems that were distributed similarly within Consents A and B (Table 1). While patients randomized
774
D~vto L. S~MELand JOHNR. FEU~SNEX
Table 1. Presenting features of enrolled patients by consent form Patient Characteristic l Age (yr, median) Education (yr, median) Race (% black) Employment status (% employed) Family member present (%) Presenting problem Paint Pruritis/skin condition Dyspnea/cough Hypertension Prostatitis/hematuria Gastrointestinal$ Ear/nose/throat problem Miscellaneous#
Consent A (n = 52)
Consent B (n = 48)
58 12 58
51 10 42
33 2
31 6
29 8 3 1 1 4 4 2
28 4 4 5 4 1 0 2
*p-Values from Wilcoxon rank sum, x2 or Fisher exact all 30.1. TIncludes all disorders for which primary problem is pain (e.g. headache, chest, abdominal, or extremity pain). tconstipation, hematochexia, nausea. §Hypercholesterolemia, diabetes, viral syndrome.
to Consent A described more severe symptoms, median symptom scores were not significantly different after adjusting for multiple comparisons (A = 7.2 mm, B = 3.8 mm, p = 0.03; p < 0.01 required for statistical significance given 5 comparisons). The baseline demographic data remained comparable after patients were partitioned according to their consent preference (n = 55 consenting, n = 45 declining). We first analyzed outcomes dealing with the primary study aims concerning the consent form and citation of quantitative information. Patients randomized to consent A were more likely to consent than those randomized to Consent B (consent rate A = 67%, consent rate B = 42%, p < 0.01; Table 2A). However, patients who cited quantitative information were no more likely to consent to participation than those who did not cite quantitative information (cited quantitative information = 62% consent rate, did not cite quantitative information = 49% consent rate, p = 0.2; Table 2B). Next, we analyzed the raw data to determine if the main effects, type of consent form, and citation of quantitative information were associated with the dependent variable, consent rate (Table 3). Patients randomized to Consent A were more likely to consent to participate than those randomized to Consent B after controlling for citation of quantitative information (p < 0.01). However, the consent rate was strongly affected
Table 2. Impact of consent form and citation of quantitative information on subseouent consent decisions Consent Decision Consent Decline
(A)
Consent Form
55
45
Consent Decision Decline Consent
(B) Cited Quantitative Information
55
45
by an interaction between the consent form and whether the patient cited quantitative information when making a decision (interaction term fi = 3.43 for consent form = A and citation of quantitative information, x2 = 7.72, p < 0.005). Patients who cited quantitative information in their decision were highly likely to consent when randomized to Consent A (95% consented to participate) and unlikely to consent when randomized to Consent B [36% consented to participate (risk ratio = 2.6, p < O.OOl), Table 31. All the subjects who consented to participate after being offered Consent A indicated they wanted the faster medication, while subjects who declined Consent B did so usually (88%) because they did not want a medication that was only half as fast as the usual treatment (Table 4). Patients who did not cite quantitative information in their decision were equally likely to participate when randomized to Consent A (50% consented to participate) or Consent B Table 3. Interaction between consent form and citation of quantitative information on consent rate Cited Quantitative Information No Yes
Form
Consent Form Form
Randomized Controlled Trial of Informed Consent Table 4. Reason for decision given by patients who did cite quantitative information Frequency Cement A (“Twice as fast”) Patients who consented (n = 19)
19
Wanted a medicine that was twice as fast Patients who declined (n = I)
1
Wanted his usual medication Consent B (“Half as fast”) Patients who consented (n = 9)
2 1 1 1
Willing to try something new Slower is better in the long run Wanted the “challenge” Would do what the doctor wanted Doctor would not suggest anything bad Wanted all medications from the VA Willing to help the VA No particular reason
1 1 1 1
Pafients who declined (n = 16)
Wanted a faster medication Not enough known about the new medication Does not like new things
14 1 1
(48% consented to participate, risk ratio = 1.1, Table 3). The most frequent reason these patients consented to participate was their preference for newer treatments, while the most frequent reason for declining was fear of experimentation (Table 5). We were unable to detect differences in reasons because of the large number of different answers among relatively few patients. Logistic models were created to determine which variables were useful in predicting the consent rate other than randomized consent form, citation of quantitative information for decision making, and interaction between consent form and quantitative information. The patient’s age (linear and quadratic terms), years Table 5. Reason for decision given by patients who did not cite quantitative information Frequency Patients who consented (n = 27)
Prefer new “things” Willing to try anything for relief Thought it would help others No particular reasons cited Willing to try if the new medication worked Did not believe it would hurt Usual medication would not work Believed side effects less with new medication Patients
who declined (n = 28)
Did not like the idea of an experiment Usual medication seemed safer or better Symptoms not severe enough Wanted more information Prior bad experience in a drug trial Did not want any medication Did not want to wait to receive new medication Patient blind and could not see consent form Trusted physician who prescribed usual treatment No particular reason cited CEu,sn
9 8 3 2 2 1 1 1 14 5 2
1 1
1 1 1 1 1
175
of education, race, and whether additional questions were asked after presentation of the consent form added no further predictive information to the logistic model (x: = 3.18, p = 0.7). After determining the lack of effect for all these variables in the entire study population, the contribution of severity of symptoms was determined for the subset of 59 patients who used the visual analog scale to describe their symptoms. Severity of symptoms did not make a linear contribution to the model (x: = 0.16, p = 0.7). However, adding the linear and quadratic terms for analog-scaled severity of symptoms to the three-variable model (consent form, citation of quantitative information, and interaction term) demonstrated a significant effect (1: = 8.35, p = 0.015). The variables consent form, citation of quantitative information, their interaction, linear and quadratic analog score were jointly significantly associated with the consent rate (xi = 13.77, p = 0.017) and had good predictive ability for consent rate [Somer DYX= 0.5, Fig. l(A,B)]. The graphic relationship of the terms in the model demonstrates that subjects with minimal or severe symptoms were less likely to consent than those with “mid-range” symptoms. DISCUSSION
The results from our study suggested strongly that what the consent forms contained and what the patient assimilated from the consent process affected the decision to participate in a research trial. Study patients who cited the information that the experimental drug worked “twice as fast” were much more likely to consent (95% consented) than patients who cited the information that the experimental drug worked “half as fast” as usual treatment (36% consented). Patients who did not cite quantitative information had similar consent rates for Consent A vs B (consent rate A = 50%, B = 48%, relative risk for consent = 1.1). The impact of such simple, quantitative information on consent rates was impressive. We infer that adding other similarly simple quantitative information into current consent forms may be useful. Unfortunately, only 45% of our patients cited the quantitative information. In order for patients to make informed decisions, future research should evaluate better ways of enabling them to recognize and use quantitative information. The consent rates were modified significantly by the severity of symptoms measured on a
776
DAVIDL. SIMIZL and JOHNR. FE~NER
(A) Probability of Conrent
0’ 0
1
2
3
4
6
Analog
6
7
8
9
10
7
8
9
10
Score
(B) Probability of Conrent 1
0.8
0.8
0
1
2
3
4
6
8
Analog Score Fig. 1. The predicted probability of consent vs the analog score, grouped by whether or not quantitative infortuation was cited iu the decision: (A) patients who cited quantitative information and (B) patients who did not cite quantitative information; ~-Consent A, A--Consent B). Probability of consent=l/{l+exp[-(-1.19+0.05*F-0.72sQ+ 2.53 * F* Q + 0.9 + analog - 0.09 + analog*)]}. F: 1 = Consent A, 0 = Consent B. Q: 1 = cited quantitative information, 0 = Did not cite quantitative information.
visual analog scale [Fig. l(A,B)]. This relationship was non-linear: patients with minimal or intense symptoms were less likely to consent than those with moderately intense symptoms. Patients who perceived their symptoms as minimal or severe may have been “risk averse”. Patients with minimal symptoms may have believed the usual treatment was sufficient so that taking risks associated with an unknown medication were not justifiable. Subjects with severe symptoms may have believed their symptoms were so severe they would rather accept the
known usual treatment than risk the new medication, especially given the new medication’s uncertain efficacy. Inferences from this finding relate importantly to patients’ tolerance for risk, but should be validated in other groups of patients and other clinical settings. The lack of a linear trend between perceived symptom severity and consent rate supports strongly the process of informed consent by contradicting the view that severely ill patients feel coerced to try anything. We created a true experiment so that patients actually believed they were making a decision that might lead to participation in an actual trial. This study demonstrates that empiric trials of the informed consent process are feasible and can be conducted without many of the methodological limitations of previous work. Our evaluation of the informed consent process addressed the five important limitations of previous work outlined by Meisel and Roth [l]. First, we tape-recorded patient-physician interactions and submitted them for review by an attorney who was blinded to the study hypotheses. This allowed us to assess what the patient was actually told and to document that the presentation was unbiased. Second, we did not succumb to the conceptual error of only evaluating patient recall. Third, rather than employing non-patients in hypothetical situations, we used real patients who were faced with making a real medical decision. Fourth, our study should be generalizable to other non-emergent ambulatory settings by virtue of the diversity of presenting problems. Despite patient’s inferring they were participating in a trial, our study’s generalizability may be limited by our use of a clinically unlikely scenario. Finally, we avoided investigator bias pertaining to informed consent research by including only subjects who were not patients of the investigator. For methodological reasons, we enhanced the validity of our study by eliminating investigator bias. However, eliminating this bias adversely influenced our overall consent rate which was relatively low (55%). We believe that the patients’ unfamiliarity with the investigator, or their more skeptical view of the consent form contributed to our low consent rate. Because we eliminated investigator bias, we could not directly assess the influence the physician might have on obtaining consent from their own patients. We believe that physicians’ influence on the informed consent process is dramatic. Physicians can affect favorably patients’ consent
111
Randomized Controlled Trial of Informed Consent
to procedures, surgery, and clinical trials by acquiring their consent, or they can affect consent unfavorably by exerting too much influence on patients’ decisions. While our study demonstrated a significant effect on consent rate when quantitative information was cited, we also observed a broad gap between disclosure and decision-only 45% of patients cited quantitative information in their decision. Additional empirical studies should focus on ways to improve patient recognition of important components of consent and assimilation of these components into their decision making. In addition, the potentially powerful role played by the physician should be systematically evaluated in varying patient situations. The gap between disclosure and decision should not create barriers to clinical care or research; on the contrary, better understanding may serve to reduce barriers and ultimately improve consent rates for clinical research. Acknowledgement-This work was supported in part by a grant from the A. W. Mellon Foundation. APPENDIX INFORMED
REFERENCES
l. Meisel A, Roth LH. What we do and do not know
about informed consent. JAMA 1981; 246: 2473-2477. 2. Appelbaum PS, Lidz CW, Me&l A. Informed Cotmeotz Legal Theory and Clinical Practice. New York: Oxford 3 University Press; 1987. President’s Commission for the Study of Ethical ’ Problems in Medicine and Biomedical and Behavioral Research. Making Health Care De&ioos, Vol. 1. Washington, D.C.: U.S. Government Printing Office; 1982. 4. Kaplan KH. Assessing judgment. Gen Hasp Psycho1 1988; 9: 202-208. 5. Appelbaum PS, Grisso T. Assessing patient’s capacities to consent to treatment. N m J Med 1988: jl9: 1635-1638. 6. Grisso T. Evaluating competencies: forensic assessments and instruments. In: Sales BD. Ed. Perrmeetlvea la Law and Psychology, Vol. 7. New York: Plenum Press; 1986. 7. Tversky A, Kahneman D. The framing of decisions and the psychology of choice. Science 1981; 211: 453458. 8. Mazur DJ. Why the goals of informed consent are not realized: treatise on informed consent for the primary care physician. J Gen Intern Med 1988; 3: 370-380. 9. Harrell FE. The LOGIST procedure. In: SUGI Supplemental Library User’s Guide, Version 5. Cary, N.C.: SAS Institute Inc.; 1986: 269-293. A
CONSENT TO EVALUATE MEDICATION
Al00 IN RELIEVING
SYMPTOMS
Your doctors are currently evaluating you for . We are investigating the use of a new medication, AlOO, versus the usual treatment for patients with your condition. The use of either the new medication or the usual treatment has no known permanent side effects. If you agree to participate in this investigation, we will determine your ability to take the new medication from the examination and tests already performed. If the medication can be given to you safely we will decide whether you will receive the experimental medication or the usual treatment by chance. This is a common way of assessing new medications. The cost to you will be the same, whether or not you participate in the investigation. It is important that you understand the following: (1)
If you choose not to participate, your physician will provide you with the best available usual treatment.
(FOR CONSENT A): (2)
The new medication, AIOO, may work twice as fast as the usual treatment.
(FOR CONSENT B): (2) I AGREE/DO
Date
The new medication, AlOO, may work half as fast as the usual treatment. NOT AGREE (circle one or the other) to participate in this investigation:
Time
