A cost-effectiveness analysis of off-label atypical antipsychotic treatment in children and adolescents with ADHD who have failed stimulant therapy

Abstract

The objectives of this study are: (1) to estimate the expected health outcomes of atypical antipsychotics (AAPs) and other non-stimulant attention-deficit/hyperactivity disorder (ADHD) medications and (2) to evaluate the cost-effectiveness of AAPs compared to other non-stimulant ADHD medications. We used decision analysis to compare three alternatives for treating children and adolescents with ADHD who failed initial stimulant treatment: (1) AAPs, (2) a selective norepinephrine reuptake inhibitor (atomoxetine), and (3) selective α2-adrenergic agonists (clonidine and guanfacine). Probability estimates and quality-adjusted life year (QALY) weights were derived from a literature review. Cost-effectiveness was estimated using the expected health outcomes derived from the decision analysis and expected costs from the literature. The study was conducted from the third-party payer perspective, and the study period was 1 year. One-way deterministic sensitivity analysis and a Monte Carlo simulation were performed. Over the course of 1 year of ADHD pharmacotherapy, the highest QALY was for clonidine/guanfacine (expected QALY = 0.95) followed by atomoxetine (expected QALY = 0.94). Atypical antipsychotics yielded the lowest health outcome with an expected QALY of 0.84. In the cost-effectiveness analysis, the AAP strategy was dominated as it was less effective and more costly than other two strategies. Compared to clonidine/guanfacine, AAPs provided lower QALYs (0.11 QALY lost) at an additional cost of $2186 on average. Compared to atomoxetine, AAPs resulted in 0.10 QALYs lost at an additional cost of $2186. In this decision analysis model, AAPs provide lower expected health outcomes than other ADHD medications in children and adolescents who failed prior stimulant therapy. Furthermore, AAPs were not a cost-effective option.

Supplementary appendix

This section provides supplementary material for the primary paper, including a more detailed presentation of several methodological points. They should be read in conjunction with the primary paper.

Standardizing probabilities from different forms presented in literature

Probabilities of events in our decision tree were obtained in various forms from the literature. One of the typical ways of presenting the effectiveness/safety of a drug in a randomized control trial (RCT) is to use a two-by-two table. Also, many RCTs report effect size, which is calculated as the difference between the treatment group mean and the control group mean divided by pooled standard deviation [i.e., effect size = (treatment mean—control mean)/pooled SD]. However, these are rarely used in observational studies. For example, studies that assessed antipsychotic agent-associated weight gain reported the average change in body weights with standard deviation. In order to convert the different forms of probabilities into a standardized probability that takes 0 as the lowest possible value and 1 as the highest possible value, we used following methods.

• 1. Calculating the standardized probability from two-by-two table (ADHD 2012).

  The effectiveness/safety of a drug can be expressed using two-by-two table in a RCT. For example, following table is based on the result of RCT conducted by Daviss et al. (2008).

	Bradycardia	No Bradycardia
Clonidine-treated (n = 31)	7	24
Placebo (n = 30)	1	29

They reported that the probability of having bradycardia in clonidine-treated children was 22.6 % (7/31 × 100 = 22.6 %), and the probability of having bradycardia in placebo group was 3.3 % (1/30 × 100 = 3.3 %). The probability of clonidine-associated bradycardia is calculated as the proportionate increase in the probability of bradycardia resulting from clonidine treatment, which is equal to 0.854 = (0.226 − 0.033)/0.226.

• 2. Calculating the standardized probability from effect size (Tickle-Degnen 2001).

  The effect size is defined as the difference between the mean outcomes for treatment and control groups in standard deviation units. Tickle-Degnen (2001) argues that because the effect size is a standard normal deviate, we can assume a normal distribution to describe the variation of individuals’ responses around the average outcomes (Tickle-Degnen 2001). For example, if the effect size is 0.65 as shown in the guanfacine RCT study conducted by Sallee et al. (2012), the probability of effectiveness is simply the area under the standard normal curve at 0.65, which is equal to 0.627.

• 3. Calculating the standardized probability from the change in body weight.

  The effect size of a drug with respect to weight gain is calculated based on the reported body weight changes of the treatment and control groups. Once the effect size is estimated, the standardized probability is obtained using the standard normal table (Tickle-Degnen 2001).

• 4. Calculating the standardized probability from hazard ratio (Chinn 2000).

  The hazard ratio is equivalent to the odds that a patient in the treatment group reaches the endpoint first (Chinn 2000). For example, the probability of developing type II diabetes first can be derived from the odds of developing type II diabetes first, which is the probability of developing type II diabetes first divided by the probability of not developing first:

  $${\text{Hazard ratio }}\left( {\text{HR}} \right) \, = {\text{ odds }} = \, P/\left( {1 \, - \, P} \right);$$
  $$P \, = {\text{ HR}}/\left( {1 \, + {\text{ HR}}} \right)$$