Assessment of relative efficacy of citalopram and escitalopram
Of 248 records, we selected 41 reviews (Figure 1). In addition, we identified four reports from health technology assessment agencies. From the 45 selected reviews, we identified 81 full-text reports concerning potentially eligible randomized clinical trials and we selected 22 eligible randomized controlled trials (RCTs) (Figure 1). Search for trials, from 2011 to 2012, yielded one additional eligible trial [32]. No additional trial from manufacturer registries or Lundbeck's list of trials was identified.
For the 23 RCTs, 4 had published results only, 6 had unpublished results only and 13 had both published and unpublished results. The trials provided for 29 randomized comparisons: 7 between citalopram and escitalopram, 12 between escitalopram and placebo and 10 between citalopram and placebo; 3 trials provided a closed loop comparison of citalopram, escitalopram and placebo (See Sections 3 and 4, Additional file 1).
A total of 2,569, 2,412 and 2,376 participants were allocated to escitalopram, citalopram and placebo, respectively. Elderly patients were included in four trials. Outcome assessment times ranged from 4 to 12 weeks. All trials were sponsored by pharmaceutical companies, except one by the Chinese National Institute for Pharmaceutical Research, (See Section 5, Additional file 1).
Meta-analysis of head-to-head trials
Of seven identified head-to-head randomized comparisons, all showed the superiority of escitalopram over citalopram except Ou 2011 [32]. Escitalopram was associated with higher response as compared with citalopram (random-effects model, combined OR 1.60 (95% CI 1.05 to .46)) (See Section 8, Additional file 1). This combined OR would translate to a NNT of 8.5 and 9.6 patients to achieve an additional response with escitalopram compared to citalopram, when the control response rate is lower (47%) or higher (61%). Heterogeneity was considerable across trials (I² = 80%; τ² = 0.26), but mainly because of one trial, Yevtushenko 2007, which showed outlying results. The funnel plot of the seven comparisons did not reveal asymmetry; see Section 9, Additional file 1.
Concerning acceptability, the proportion of treatment completers was greater with escitalopram than citalopram (Section 10, Additional file 1). For escitalopram versus citalopram, the random-effects combined OR was 1.27 (0.93 to 1.72), with moderate heterogeneity (I² = 26% and τ² = 0.04).
Adjusted indirect comparisons of placebo-controlled trials
For the two meta-analyses of placebo-controlled comparisons of citalopram (n = 10 trials) and escitalopram (n = 12), we found no substantial heterogeneity across trials (I² = 0% and τ² = 0.00 for citalopram vs. placebo; I² = 27% and τ² = 0.02 for escitalopram vs. placebo), Section 6, Additional file 1. Patients and trial characteristics were similar for the two sets of placebo-controlled trials, see Section 6, Additional file 1.
The proportion of responders was significantly greater with citalopram and escitalopram than placebo and the two effect sizes were of similar magnitude. Random-effects combined OR 1.50 (1.27 to 1.78) for citalopram and 1.55 (1.33 to 1.82) for escitalopram. From these estimates, the adjusted indirect comparison OR for citalopram versus escitalopram was 1.03 (0.82 to 1.30). We found a large inconsistency between the direct and indirect estimates (difference in log ORs 0.44, corresponding to a ratio of OR of 1.55, P = 0.07, Figure 2. Consequently, we could not combine the direct and indirect estimates in a network meta-analysis. Moreover, we could not perform a NNT analysis because of a lack of difference in efficacy between citalopram and escitalopram.
For each set of placebo-controlled trials, we found no evidence of small-study effect (See Section 12, Additional file 1; Egger's test P = 0.79 for citalopram vs. placebo and P = 0.46 for escitalopram vs. placebo).
Concerning acceptability, the proportion of treatment completers was lower with citalopram and escitalopram than placebo, with similar effect sizes, see Section 13, Additional file 1, the random-effects combined OR 0.91 (0.75 to 1.10) for citalopram and 0.89 (0.73 to 1.07) for escitalopram. From these estimates, the adjusted indirect comparison OR for escitalopram versus citalopram was 0.98 (0.75 to 1.28), with large inconsistency between the direct and indirect estimates (difference in log ORs 0.26, corresponding to a ratio of OR of 1.30 (P = 0.21), Figure 2. We found no important heterogeneity in the two sets of trials (I² = 0% and τ² = 0.00 for citalopram vs. placebo; I² = 25% and τ² = 0.03 for escitalopram vs. placebo).
The sensitivity analyses of the direct and indirect comparisons after excluding trials without comparable dosages, with imputed outcome data or of older adults only gave results consistent with those from the primary analyses (data not shown).
Assessment of reimbursement costs for citalopram, its generic drugs and escitalopram
Sections 7 and 14, Additional file 1 show the evolution of the number of claims for each drug form. The projected results from the EGB sample showed a substantial decrease in consumption of citalopram between 2004 (2.1 million claims) and 2006 (0.7 million claims). This decrease was accompanied by an increase of approximately twice the claims for the generic forms of citalopram during the same period. Moreover, the new revised-formula escitalopram represented 40% of the market share in 2006 (1.7 million claims) after it was introduced to the market in April 2005. Between 2006 and 2011, claims for citalopram continued to decrease, to a lesser extent, and that for the generic forms continued to increase, to peak in 2008, which was followed by a slight decrease up to 2010. However, escitalopram claims grew even more steeply towards the end of 2010, reaching 5.4 million claims. By the end of 2010, escitalopram consumption had exceeded that of citalopram and its generic forms combined (5.4 million claims for escitalopram vs. 0.2 and 1.7 million for citalopram and its generic forms, respectively) (Sections 7, 16 and 17, Additional file 1).
Consumption in DDD units showed changes similar to reimbursement results; for citalopram consumption, the DDD units decreased from 73.9 million in 2004 to 7.6 million in 2010, whereas for escitalopram, the units increased from 15.7 million in 2005 to 193.9 million in 2010. For generic forms of citalopram, the DDD units were 55.2 million in 2005 and slightly increased to 58.8 million in 2010 (Sections 7 and 15, Additional file 1).
Reimbursement costs reflected the trends in consumption (Figure 3; Section 7, Additional file 1). The total monthly cost for the drugs was 5.6 million Euros when the generic forms were introduced into the market. Although the total monthly cost slightly decreased to 5.2 million Euros when escitalopram was introduced into the French market, in May 2005, the monthly cost reached 6.1 million Euros a year subsequently (Figure 4). The cost burden of escitalopram continued to increase, to reach 96.8 million Euros in 2010, as compared with citalopram, 4.4 million Euros (Figure 3). For the generic forms of citalopram, the cost was >20 million Euros from 2005 to 2010. Moreover, the reimbursement cost for escitalopram exceeded that of citalopram and its generic forms combined (Figure 5). Overall, the health cost burden of the three drug forms reached 120.6 million Euros in 2010 (see Figure 5 and Section 17, Additional file 1).