Multicentre randomised control trial comparing real time teledermatology with conventional outpatient dermatological care: societal cost-benefit analysis

Dear Sir,

The article by Wootton et al. is a valuable addition to the limited
literature on the cost effectiveness of telemedicine applications.
However, we have some reservations about their calculations of the net
societal cost of a teledermatology consultation.

In their study the authors are comparing outpatient dermatology,
current clinical practice, with real time teledermatology in terms of
clinical outcomes, cost-benefit (which should theoretically include the
monetary valuation of clinical outcomes) and patient satisfaction.

Having calculated a total cost of £201.88 for a teledermatology
consultation the authors subtracted two values from this total to arrive
at the net societal cost. First, and quite correctly, they deducted the
"savings" that would be made from reduced dermatology referrals, which
they attribute to the "learning benefits and increased confidence in
managing patients obtained from the joint videolink consultation".

Our point of contention is that they were wrong to subtract a further
£69.78 for the "benefits" of the programme. These "benefits" do not
reflect clinical outcomes, as the authors found no major differences in
the two approaches in terms of clinical outcomes. Rather it is an imputed
measure of the training cost necessary to achieve the same educational
impact as the experience obtained using telemedicine. This would,
presumably, lead to a similar reduction in dermatology referrals as
achieved by telemedicine. This then begs the question of who exactly in
society would actually benefit to the tune of £69.78, as this training is
not currently provided in current clinical practice.

This result, as presented in the paper, appears to be a form of
double counting. Real time teledermatology may or may not be cost
effective but its societal benefits surely cannot include the savings from
reduced referrals AND the cost of training necessary to obtain a similar
pattern of GP referrals in the absence of telemedicine.

Yours faithfully,

Paul Jacklin & Jenny Roberts

Comparison of a telemedicine application to a conventional medical
approach using a randomized trial design is uncommon in the literature.
Thus, the report by Wootton et al. is a major contribution to the
evaluation of telemedicine. However, their report raises several
methodological questions.

One issue concerns the idea of the episode of care. That is, what is
the total impact of the telemedicine application the treatment of a
disease or condition from diagnosis to resolution. This concept has
several implications for the study design. First, the report is unclear
as to whether or not the patients were randomized at the time of diagnosis
of an incident disease or condition or if the disease or condition had
worsened to the point where the general practitioner felt a call to a
specialist were required. If the latter were true, incident versus
prevalent case is a potential confounder and should be controlled for.

Second, all subsequent visits should contribute to the treatment
group to which a patient was assigned, i.e., intent to treat. In Table 2,
all conventional visits were assigned to the conventional arm even though
many of these were for patients randomized to the telemedicine arm. If
physicians took substantially longer for a conventional visit when the
patient had previously had a telemedicine consult, this would not be
apparent in this study design.

Third, economic data collection was anonymous. Thus, it is
impossible to determine if there are differences among the costs of an
episode of care for any given patient.

Fourth, the study does not seem to target the episode of care. That
is, there is no attempt to determine if the disease or condition resolved
as a result of treatment. A table of the diseases or conditions for
patients within this trial would be useful in this regard. The "clinical"
outcomes used here (health care system interactions) are economic outcomes
and not clinical outcomes. A better clinical outcome is whether or not
there was a resolution of the disease or condition. For those patients
whose disease or condition did not resolve within the time frame of the
trial, the costs for episode of care can be censored for purposes of
analyses (see, for example, Gray).

There are other concerns with this report. The power calculation
does not describe a particular variable and the magnitude of difference
anticipated for this variable. This is a cost-benefit study although the
authors occasionally refer to cost-effectiveness. No appropriate
effectiveness measure (such as the number of cases resolved) is included
in this report. Thus, it is not clear that the sample size is
appropriate.

Also, adverse events are not specifically considered by these
investigators. One wonders what the impact of a rare (less than 1% of
cases) but expensive adverse event (such as an incorrect diagnosis) might
be on the economic analyses. This relates directly to the power question.
For example, a sample size of 100 per group has 80% power to detect a
difference between an event with a rate of 1% in one group and 10% in a
second group. Smaller differences between the groups would require much
larger sample sizes (such as 2,320 per group to detect a doubling in the
rate from 1% to 2%).

In the Discussion, the authors equate a patient's failure to return
for a recommended general practice follow-up visit with clinical
effectiveness of the consultation. This assumption seems too strong. For
example, the original referral may have been unnecessary to begin with or
the patient may be generally non-compliant with a physician's
recommendations. If the recommendation made by the dermatologist for a
general practice follow- up visit was unnecessary, this would be
clinically ineffective. By the authors' logic, if the patient failed to
comply with the recommended follow-up visit, say for example, because the
patient was killed in a accident, then the consultation was effective.

Other potential confounding variables such as urban versus rural,
age, and severity of condition could have controlled for in the analysis,
either by stratification or regression analyses. The response rate of
about 60% is sufficiently low as to be worrisome. Non-anonymous economic
questionnaires would have allowed the researchers to send second notices
to patients who failed to respond. The potential validity problems with
patient-reported times should be discussed.

This clinical trial and economic analysis supports the concept that,
given an adequate volume of patients, care via telemedicine can be as or
less expensive that conventional care in some cases. This study does not
adequately address the question of whether or not the outcome of an
episode of care (including adverse events) would be the same for both
approaches.

Dale F. Kraemer

Benjamin K. Chan

William Hersh

Division of Medical Informatics and Outcomes Research
School of Medicine
Oregon Health Sciences University
Portland, Oregon, USA