Table 3.

Transition probabilities used in the Markov process that do not vary by patient population

			Probability by treatment
		Atomoxetine	IR-MPH	XR-MPH	No medication
Probability of one or more medication-related adverse eventsa		0.129	0.129	0.129	0.000
Probability that a medication-related adverse event is insomniab		0.000	0.48	0.48	NA
Probability that a medication-related adverse event, which is not insomniac	First 4 cycles	0.473	0.473	0.473	NA
	Cycles thereafter	1.000	1.000	1.000	NA
Probability that insomnia will persist from one Markov cycle to the nextd	First 4 cycles	NA	0.953	0.953	NA
	Cycles thereafter	NA	1.000	1.000	NA
Probability that a non-responder discontinues due to lack of efficacy during a Markov cyclee		0.0989	0.0989	0.0989	NA
Probability that a patient discontinue due to a medication-related adverse event during a Markov cyclef		0.1209	0.1209	0.1209	NA
Probability that a patient discontinues for reasons other than lack of efficacy or a medication-related adverse event during a Markov cyclee	First 4 cycles	0.384	0.384	0.384	NA
	Cycles thereafter	0.000	0.000	0.000	NA

Abbreviations: IR-MPH = immediate-release methylphenidate; XR-MPH = extended-release methylphenidate; NA = Not applicable

a. Probabilities based on post hoc analyses of safety data pooled from six randomised placebo-controlled trials of atomoxetine versus placebo [38-40,42] (some are data on file). Assumption of parity between active treatments based on similar post hoc analyses of data from a limited open-label direct comparator study [46], supported by data from a double-blind randomised trial of atomoxetine and XR-MPH (data on file) where the proportions of patients experiencing one or more adverse events of any nature were not significantly different between the active treatments. Values are net of the placebo rate, meaning that the 'no medication' probability is zero, by definition.

b. The probability based on the relative risk (0.417) of insomnia (atomoxetine vs IR-MPH), estimated in an indirect meta-analysis of safety data [43], applied to the risk of insomnia for atomoxetine (4.7%) derived from pooled analysis of safety data from six pivotal randomised placebo-controlled trials of atomoxetine [38-40,42] (some are data on file), giving a rate of insomnia for IR-MPH of 4.7/0.417 = 11.27%. The model assumes that insomnia is experienced only as a result of taking medication. Therefore, the probability for placebo is not applicable (i.e. zero) and the probabilities for active treatments are net of the placebo rate (i.e. subtract 5.1%). As a consequence, the model assumes that patients on atomoxetine have no risk of medication-related insomnia. Patients on IR-MPH who experience insomnia will come only from the population who experience one or more adverse events as derived in note 1, therefore, for 'if adverse event, probability that insomnia included' = (11.27-5.1)/12.9 = 48%. Parity is assumed between IR-MPH and XR-MPH [24,26].

c. Probabilities based on temporal course of treatment-emergent adverse events (data on file) where weekly reports from patients treated with atomoxetine over 52 weeks imply that, for most patients, medication-related averse events mainly occur early in the treatment and are likely to resolve within approximately 16 weeks. The probability of 0.473 (0.05^1/4) for the first four cycles with adverse event(s) reflects a nominal 5% of patients in whom adverse events (that are not insomnia) persists over this duration of the Markov process. The duration of persistence of adverse events (that are not insomnia) is assumed to be similar for each medication.

d. Probabilities based on a survey of six consultant child and adolescent psychiatrists (data on file). Responses suggested that 82.5% of cases of stimulant-related insomnia would persist for more than 16 weeks. The model assumes that patients with stimulant-related insomnia that persists beyond four cycles will continue to have insomnia as long as they remain on treatment. The probabilities of 0.953 (0.824^1/4) for the first four cycles of the Markov process and 1.000 for cycles thereafter reflect this.

e. Probabilities based on discontinuation rates, regardless of treatment, from data pooled from seven randomised placebo-controlled trials of atomoxetine [38-42] (some are data on file), adjusted for differences between trials with respect to duration of follow-up. Discontinuations due to lack of efficacy were assumed to occur in only the non-responder population. In each case, parity is assumed between the active treatments.

f. Probabilities based on discontinuation rates due to adverse events from data pooled from six pivotal randomised placebo-controlled trials of atomoxetine [38-40,42] (some are data on file), adjusted for differences between trials with respect to duration of follow-up. Discontinuations due to adverse events were assumed to occur only in the population experiencing one or more medication-related adverse events and therefore were net of the placebo rate. In each case, parity is assumed between the active treatments.