Kennedy ADM, Sculpher MJ, Coulter A, Dwyer N, Rees M, Abrams KR, Horsley S, Cowley D, Kidson C, Kirwin C, Naish C, Stirrat G. Effects of Decision Aids for Menorrhagia on Treatment Choices, Health Outcomes, and CostsA Randomized Controlled Trial. JAMA. 2002;288(21):2701-2708. doi:10.1001/jama.288.21.2701
Author Affiliations: Health Economics Research Group, Brunel University, Uxbridge, England (Mr Kennedy); Centre for Health Economics, University of York, York, England (Dr Sculpher); Picker Institute Europe, Oxford, England (Dr Coulter); Weston General Hospital, Weston-super-Mare, England (Dr Dwyer); John Radcliffe Hospital, Oxford, England (Dr Rees and Ms Naish); Department of Epidemiology and Public Health, University of Leicester, Leicester, England (Dr Abrams); Princess Margaret Rose Hospital, Swindon, England (Ms Horsley); St Michael's Hospital, Bristol, England (Ms Cowley and Dr Stirrat); Taunton and Somerset Hospital, Taunton, England (Ms Kidson); and St Paul's Hospital, Cheltenham, England (Ms Kirwin).
Context Decision aids can increase patient involvement in treatment decision
making. However, questions remain regarding their effects and cost implications.
Objective To evaluate the effects of information, with and without a structured
preference elicitation interview, on treatment choices, health outcomes, and
Design, Setting, and Participants A randomized controlled trial with 2 years of follow-up. Between October
1996 and February 1998, 894 women with uncomplicated menorrhagia were recruited
from 6 hospitals in southwest England. Women were randomized to the control
group, information alone group (information), or information plus interview
Interventions Women in both intervention groups were sent an information pack (a booklet
and complementary videotape) 6 weeks before their specialist consultation.
Immediately before their consultation, women in the interview group underwent
structured interview, to clarify and elicit their preferences.
Main Outcome Measures Self-reported health status was the main outcome; secondary outcomes
included treatments received and costs. Cost analyses adopted a UK health
service (payer) perspective, and were based on patient-reported resource use
data and are reported in 1999-2000 US dollars.
Results The interventions had no consistent effect on health status. Hysterectomy
rates were lower for women in the interview group (38%) (adjusted odds ratio
[OR], 0.60; 95% confidence interval [CI], 0.38-0.96) than in the control group
(48%) and women who received the information alone (48%) (adjusted OR, 0.52;
95% CI, 0.33-0.82). The interview group had lower mean costs ($1566) than
the control group ($2751) (mean difference, $1184; 95% CI, $684-$2110) and
the information group $2026 (mean difference, $461; 95% CI, $236-$696).
Conclusions Neither intervention had an effect on health status. Providing women
with information alone did not affect treatment choices; however, the addition
of an interview to clarify values and elicit preferences had a significant
effect on women's management and resulted in reduced costs.
The call for increased patient participation in treatment decision making
has come from a range of different perspectives.1- 12 Methods
to achieve these aims have included the development of patient decision aids.
These can provide evidence-based information on treatment options and outcomes,
help patients consider the personal value they place on benefits vs harms,
and participate in decisions about their care. Decision aids come in a variety
of formats including leaflets, audiotapes, decision boards, computer programs,
videos, Web sites, and structured interviews.13,14 Systematic
reviews have shown that they increase patients' knowledge of available management
choices without increasing anxiety or decisional conflict.13 They
reduce the number of patients who are uncertain about what to do and assist
in promoting participative decision making.13 However,
the effect on treatment choices, health outcomes, satisfaction, and costs
is less understood.
The clinical context of this study is menorrhagia, or heavy menstrual
bleeding. Around a third of all menstruating women report heavy periods,15 and it is a common reason for consultation with a
family physician.16 Referral to specialist
care is common, with 35% of those consulting their family physician receiving
a referral within a year.17 The condition is
not life threatening, but it can have a considerable effect on a woman's quality
A woman requiring treatment for menorrhagia faces a number of treatment
options, including provision of advice and reassurance, addressing possible
iatrogenic causes, drug therapy, or referral to a gynecologist for possible
surgery, typically hysterectomy or endometrial destruction.19 Women
are faced with clear trade-offs between treatments, and their preferences
must be taken into account in decision making.8,10,20 Decision
aids were developed to help women with menorrhagia determine and articulate
their treatment preferences, and to enable them to play a more active role
in treatment decision making. The aids used in this study were: (1) an information
pack consisting of a specially designed booklet and linear video; and (2)
a structured interview with a research nurse to clarify and elicit a woman's
treatment preferences. This study evaluated these aids in terms of their effects
on outcomes, patient management, and costs.
The content of the information pack drew on epidemiological studies,
a systematic review of treatment efficacy published in the Effective Health Care series,19 and
research into patients' information needs, including 4 specially commissioned
focus groups.21 The 28-page booklet included
an introductory section emphasizing the importance of a woman's preference
in deciding on treatment, together with chapters describing menorrhagia and
its causes, investigations, treatment options (medical and surgical), and
the benefits and risks of surgery. It also included a section entitled "personal
treatment plan" in which the reader was prompted to write down her preferences
in response to a series of questions. The 30-minute video was presented by
a female physician and included clips of interviews with women who had experienced
different treatments for menorrhagia. The materials were pilot tested with
patients before final production.
The structured interview drew on material derived from patient surveys22,23 and interviews with women with menorrhagia.21 It was comprehensively piloted with 32 women during
the development phase. The purpose of the interview was to help patients clarify
and articulate their preferences, and to give them the chance to provide information
that they might not have the opportunity or inclination to reveal to their
physician. The interviews were conducted by a trained research nurse immediately
before the woman's first specialist consultation for menorrhagia. Information
on the woman's attitudes toward the clinical and lifestyle characteristics
of possible treatments, the extent to which she wanted to become involved
in the decision-making process, and her treatment preference were summarized
on a form. The woman was encouraged to give the form to her physician during
the consultation, after which it was added to her medical record.
A total of 28 consultant gynecologists from 6 hospitals in southwest
England took part in the study. Institutional review board approval was granted
by each of the centers. All women who were referred from primary to secondary
care with uncomplicated menorrhagia, which was deemed nonurgent by their consultant,
were considered for trial entry if their referral related to a new episode
of menorrhagia. Women were identified by a research nurse from inspection
of referral letters sent from general practitioners to the participating consultants.
All women identified from referral letters were registered and written consent
For those consenting, random allocation to 1 of the 3 groups was then
performed using a form of random permuted blocks, with block size randomly
set to 3, 6, or 9 to avoid any possibility of selection bias.24 The
allocation sequence was generated by computer and stratified by consultant
and the age at which the woman left full-time education. Secure randomization
was ensured by using a central telephone randomization system. It was not
possible to blind women or their physicians to allocation group. Women were
randomized to (1) a standard practice control group with no intervention;
(2) an information group, which received the booklet and videotape, sent to
their homes 6 weeks before their consultations; and (3) an interview group,
which received the same materials as the information group and who underwent
an interview immediately before their consultation. The interview attempted
to explicitly clarify values and elicit treatment preferences.
Before randomization, participants completed a baseline questionnaire,
which included generic and condition-specific health status measures, and
elicited clinical and sociodemographic characteristics, treatment knowledge,
and treatment preferences. Women randomized to an intervention arm were asked
to complete a questionnaire about their use of the information pack. Women
in the interview group were asked to arrive 30 minutes before their scheduled
outpatient appointment to participate in the interview. Follow-up questionnaires
were sent at 6, 12, and 24 months postconsultation. The questionnaires focused
on outcomes and health service contacts. Women who did not respond to the
24-month questionnaire after receiving 2 reminder letters were asked to take
part in a short telephone interview covering key items from the follow-up
Outcomes were measured using standard validated scales when available.
The primary outcome was general health status measured using the SF-36 (36-item
Short Form Health Survey).25 It was considered
important to use a generic health status measure to permit comparisons with
other uses of health care resources. Secondary outcomes included treatments
received during follow-up, severity of menorrhagia,26 and
Nonresponse bias was assessed using t and χ2 tests. Multiple regression methods were used to analyze health status,
logistic regression to analyze treatments, and ordinal regression to analyze
satisfaction. To allow for potential clustering, both consultant (when methods
were available) the consultant-intervention interaction were modeled as random
effects.27 A pool of covariates, selected from
the baseline characteristics and treatment preferences summarized in Table 1, were defined for each analytical
model a priori.
The primary outcome of the study was health status as measured by the
SF-36. The sample size of 900 participants was calculated to detect differences
between study groups of 5 points, with a power of 80% at the .05 significance
level. This applied to all domains except the 2 role-related scales, which
have SDs more than a third greater than the other scales.28
For the cost analysis, unit costs were estimated from published sources.29- 33 We
took a UK National Health Service perspective based on 1999-2000 UK prices
and converted them to US dollars using the average exchange rate for 2000
(£1 = $1.52). For purposes of costing, we recorded resources associated
with the development and production of the interventions; duration of time
devoted by the nurse to the interview; and a woman's use of health services
(therapeutic and diagnostic procedures and medications for menorrhagia; inpatient
days in hospital for any reason; and outpatient and family physician visits
for any reason during follow-up). These data were collected using questionnaires
sent to women at 6, 12, and 24 months.
Given that the time horizon of the analysis was only 2 years, total
costs are not discounted. To account for the skewed nature of the data, 95%
confidence intervals (CIs) for differential costs have been calculated using
bias-corrected and accelerated nonparametric bootstrapping (based on the 25th
and 75th percentiles).34 For some patients,
resource use data were wholly or partially missing. We have addressed this
problem using multivariate multiple imputation methods to impute the missing
observations for these patients, under the assumption that these data were
missing at random.35
Between October 1996 and February 1998, 1301 eligible women were invited
to partipate in the study—894 (69%) gave their consent (Figure 1). There were no exclusions after randomization. Recruitment
at the 5 main hospitals ranged from 147 to 202 women; a further 48 women were
recruited from a smaller hospital that joined the trial during the first year
of recruitment. There was no significant difference in age between those granting
(mean [SD], 40 [7.0] years) and refusing (41 [7.7] years) consent (P = .56).
Table 1 gives baseline descriptive
statistics for the 3 groups. The educational background of the sample was
close to population averages.36 The 3 groups
were generally well matched but there were small between-group differences
in the duration of menorrhagia, and also in previous experience of hormonal
and nonhormonal drug treatments. For women's treatment preferences, there
were differences between the groups in terms of whether a preference was held
and whether there was a preference for individual treatments. Few women expressed
a preference for endometrial destruction, unspecified surgery, no treatment,
or any other possible treatment.
Information pack questionnaires were returned by 519 (87%) women in
the intervention groups. All but 4 women reported watching or reading at least
some of the video or booklet. The interview, which lasted a mean (SD) of 20
(6.2) minutes, was conducted with 240 (80%) of the women randomized to that
group. Clinicians' perceptions of consultation length differed between the
groups (P = .01). The clinicians perceived consultations
with women in the interview group to be "longer than usual" (28.5%) more often
than those for women in the other 2 groups (18.9% for control and 16.9% for
The response rate to the 2-year follow-up was 70% (Figure 1) with a mean duration of follow-up of 26 months for each
group. The rates of loss to follow-up varied between the groups, but these
differences were not statistically significant (P =
.69). Nonresponders were significantly younger than responders at baseline
(mean [SD] age: 39 [7.8] vs 41 [6.5] years; P<.001);
had more cases of severe menorrhagia (50 [15.7] vs 46 [13.8]; P = .001); had a lower level of knowledge (63 [21.4] vs 68 [21.0]; P = .002), and were less likely to have had previous treatment
with nonhormonal drugs (66 [25.0%] vs 241 [38.8%]; P<.005).
There were no between-group differences.
The role physical dimension of the SF-36 shows a significant difference
between the interview and control groups (P = .04).
None of the other differences reached statistical significance. Figure 2 shows the adjusted mean health status scores for the 3
groups, and illustrates the lack of a consistent effect across the dimensions.
However, the considerable improvement in health status from baseline for all
3 groups is evident.
Treatment data were reported by 81% of women during the 2 years of follow-up.
This rate was similar for all 3 groups (82% for control; 78% for information;
and 84% for interview) (P = .17). The treatments
reported by women are summarized in Table
2. Among women who underwent treatment, the interview group showed
significant reductions in hysterectomy rate in comparison with the control
group (P = .04) and the information group (adjusted
odds [OR], 0.52; 95% CI, 0.33-0.82; P = .008) equivalent
to relative reductions of 26% and 32%, respectively.37 There
was no difference between the information and control groups (P = .53). No other treatments showed a significant between-group difference,
although the numbers of women who received endometrial destruction or other
treatment were too small to rule out an effect. The intervention groups may
have been more likely to undergo drug therapy, but the differences did not
reach statistical significance (information vs control: P = .17; interview vs control: P = .11). The
observed difference in hysterectomy rates between the interview group and
the others was maintained when women reporting that they were waiting for
hysterectomy were included in the analysis (interview vs control: P = .008; information vs interview: adjusted OR, 0.52; 95% CI, 0.34-0.80; P = .005).
The interview group rated both the opportunities they had been given
to take part in treatment decision making (adjusted OR, 1.49; 95% CI, 1.11-2.01; P = .008) and the overall results of their treatments (adjusted
OR, 1.44; 95% CI, 1.03-2.01; P = .03) significantly
higher than the control group did. The differences between the information
group and the controls were smaller and not significant (opportunities to
take part in decision making: adjusted OR, 1.24 [95% CI, 0.91-1.69]; results
of treatment: adjusted OR, 1.16 [95% CI, 0.85-1.60]). The differences between
the intervention groups were not statistically significant.
Resource use is reported in Table
3. Total fixed intervention costs for the development and production
of the booklet and videotape (including expert input) came to $121 904,
which gave a cost per patient of $0.20. This assumed that the effective life
of the interventions was 3 years, and the population for whom the interventions
would be applicable was all women in England and Wales aged 25 to 52 years,
who are referred from primary to secondary care with uncomplicated menorrhagia.17,38 A total cost per patient of $21 for
information and $27 for interview included the costs of the booklet and videotape
and 20 minutes of an experienced nurse's time for the interview.39
Table 4 presents the mean
costs and the mean between-group differences in these costs. The main factors
that influenced costs were surgery and other procedures, and inpatient and
outpatient costs. Lower surgery costs in the interview group came from the
reduction in hysterectomy rates, but this was partially offset by the greater
number who received endometrial destruction and the levonorgestrel-releasing
intrauterine contraceptive device. The higher rates of nongynecology health
service contacts had the greatest influence over the between-group differences
in inpatient and outpatient costs.
In terms of overall costs, both intervention groups showed major mean
cost savings in comparison with the control group. The interview group also
showed a savings in comparison with the information group. However, the overall
cost was greatly influenced by the number of hospital contacts.
Overall costs were sensitive to the costs associated with health service
contacts and were particularly high for a small number of women with high
levels of resource use (Table 3).
However, even when unrelated and inpatient costs were excluded, interview
remained less costly than the other 2 groups (Table 4). Further sensitivity analyses were undertaken to explore
the implications of varying the fixed costs of the interventions and the cost
implications of the perceived additional length of the initial specialist
consultation for women in the interview group. Halving the population for
whom the interventions would be applicable, reducing the effective life of
the interventions from 3 years to 1 year, and increasing the length of the
initial consultation by 50% did not affect the conclusions of the cost analysis.
Among studies evaluating decision aids, this is the largest, with the
longest period of follow-up. This has enabled a number of questions to be
addressed that previous studies have been unable to answer. These include
the effects on health status, satisfaction, and long-term effects on treatments.
In addition, few studies have examined the effect of decision aids on costs.
A potential limitation of the study design is the possibility of contamination
bias, in which clinicians could have applied the experience gained from consultations
with the intervention groups in their consultations with the control group.
This bias would have the effect of reducing any observed differences in outcome
between the intervention groups and the control group, so our conclusions
may be considered conservative. Another bias could have been introduced if
there was a clustering effect in terms of the consultation style of particular
consultants or the types of patients referred to them. This has been dealt
with by incorporating consultants as a random effect in the statistical analysis.
A further potential bias could have resulted from the inability to ensure
that clinicians were blind to the allocation group. They were not told which
patients were randomized to the control or information only groups, but we
could not be certain that patients did not mention this fact in the subsequent
consultation. Patients in the interview group were given a summary sheet to
take into the consultation, so clinicians would have been aware of the allocation
in this case.
The rate of loss to follow-up is within the range of other studies in
the field,40- 42 although
these studies cover just 1 year of follow-up. There were differences in baseline
characteristics between responders and nonresponders. The statistical analyses
ensure that these differences will not affect the comparisons between trial
groups, but they should be noted when attempting to generalize the findings.
Neither intervention had a major impact on health status. The role physical
dimension of the SF-36 did differ between the interview and control groups,
but there was no consistent differential effect across the other dimensions,
and this finding is consistent with previous studies.43 Considering
the major improvements in health status from baseline levels for women in
all 3 groups, the impact of the decision aid is likely to be less important
than the effectiveness of the available treatments.
In terms of the effects on treatment received, women who received the
interview (plus information) were considerably less likely to undergo hysterectomy
in comparison with those who received standard care and those who received
information alone. This study confirms the effect identified in a recently
updated systematic review.43 Five studies showed
reduced relative rates of surgery (ranging from 21% to 42%) and a nonsignificant
trend when decision aids were compared with standard practice or less complex
decision aids. These studies concerned decisions relating to treatment for
benign prostatic hyperplasia,40 ischemic heart
disease,44,45 low back pain,41 and breast cancer.46 As
in this study, all these decisions relate to conditions with a choice between
radical surgery and more conservative options, but the results both support
and contrast the findings reported in this article. They are consistent with
the reduction in the rate of radical surgery reported here, and the additional
effect of complex decision aids in comparison with simple decision aids. However,
the effect sizes reported across the studies remain similar irrespective of
the content of the interventions, which range from relatively simple to complex
decision aids. A major strength of this article is that these effects have
been compared within the same study.
The effect of the interview on satisfaction, in which women were more
satisfied with the opportunities they had been given to be involved in making
treatment decisions, is in contrast with results in the literature.43 No effect was detected in 4 of 5 studies measuring
satisfaction with the decision-making process, or 5 of 6 studies looking at
satisfaction with the decision. It should be noted, however, that these earlier
studies had limited power to detect between-group differences in satisfaction.
The results of the cost analysis suggest that information plus interview
has lower costs than information alone and standard practice within secondary
care treatment of menorrhagia. The study indicates important reductions in
hysterectomy and inpatient and outpatient costs. Even if the cost of lowered
inpatient days and outpatient and family physician visits unrelated to menorrhagia,
which are heavily influenced by a small number of women with high costs, is
excluded from the analysis, interview remains a less costly intervention compared
with both control and information alone.
The model of information and interview used in this study could easily
be adapted for use in other settings. However, the interventions were aimed
at patients and did not involve their physicians. Future decision aids should
incorporate training for clinicians in the principles and practice of involving
patients in treatment decision making, alongside evidence-based information,
value clarification, and preference elicitation for patients. Further research
is required into which aspects of the interview contribute to the observed
effects; when in the decision-making process such interventions should be
integrated; and in which clinical settings these methods would be the most
effective and the most cost-effective.
In conclusion, providing women with evidence-based information alone
did not affect health outcomes, satisfaction, or treatments, although it may
have lowered costs. However, the addition of a structured interview helped
women to use the information to clarify their values and preferences. These
were summarized on a pro forma and given to the patients to give to their
clinicians to expedite the consultation and for inclusion in their medical
record. This process reduced costs and had a significant effect on women's
subsequent management and long-term satisfaction.