Analysing time to event data in dementia prevention trials: The example of the guidage study of EGB761®

Abstract

Time-to-event analysis is frequently used in medical research to investigate potential diseasemodifying treatments in neurodegenerative diseases. Potential treatment effects are generally evaluated using the logrank test, which has optimal power and sensitivity when the treatment effect (hazard ratio) is constant over time. However, there is generally no prior information as to how the hazard ratio for the event of interest actually evolves. In these cases, the logrank test is not necessarily the most appropriate to use. When the hazard ratio is expected to decrease or increase over time, alternative statistical tests such as the Fleming-Harrington test, provide a better sensitivity. An example of this comes from a large, five-year randomised, placebo-controlled prevention trial (GuidAge) in 2854 community-based subjects making spontaneous memory complaints to their family physicians, which evaluated whether treatment with EGb761® can modify the risk of developing AD. The primary outcome measure was the time to conversion from memory complaint to Alzheimer's type dementia. Although there was no significant difference in the hazard function of conversion between the two treatment groups according to the preplanned logrank test, a significant treatment-by-time interaction for the incidence of AD was observed in a protocol-specified subgroup analysis, suggesting that the hazard ratio is not constant over time. For this reason, additional post hoc analyses were performed using the Fleming-Harrington test to evaluate whether there was a signal of a late effect of EGb761®. Applying the Fleming-Harrington test, the hazard function for conversion to dementia in the placebo group was significantly different from that in the EGb761® treatment group (p = 0.0054), suggesting a late effect of EGb761®. Since this was a post hoc analysis, no definitive conclusions can be drawn as to the effectiveness of the treatment. This post hoc analysis illustrates the interest of performing another randomised clinical trial of EGb761® explicitly testing the hypothesis of a late treatment effect, as well as of using of better adapted statistical approaches for long term preventive trials when it is expected that prevention cannot have an immediate effect but rather a delayed effect that increases over time.

Introduction

Time-to-event analysis is frequently used in medical research to investigate the influence of potential risk factors or protective factors, and notably treatments, on the occurrence of binary clinical outcomes. One area of medicine where this type of analysis is widely used is for evaluating potential disease-modifying treatments in neurodegenerative diseases, where the therapeutic goal is to slow down the progression of loss of function. Examples include evaluating treatments that improve survival in amyotrophic lateral sclerosis (1, 2), treatments that delay initiation of L-DOPA therapy in Parkinson’s disease (3) and treatments that delay the diagnosis of Alzheimer’s disease (AD) in patients with age-related cognitive impairment (4., 5., 6.).

In these clinical trials, potential treatment effects are generally evaluated using the logrank test. However, there is generally no prior information as to how the hazard ratio for the event of interest actually evolves over time, and it is possible that the treatment effect is not constant over time, either waning as the disease progresses or increasing over time and often appearing after a certain lag time (delayed onset of action). In these cases, the logrank test has a reduced sensitivity to detect treatment effects and is not necessarily the most appropriate to use. When a decreasing, transient, increasing or delayed effect is expected or suspected alternative statistical tests are justified (7). These include the family of Fleming-Harrington tests, where the contribution of each event of interest to the overall treatment effect may vary over time according to the value of two parameters p and q. The logrank test is equivalent to a Fleming-Harrington test with p = 0 and q = 0. A value of p larger than 1 makes the test more sensitive to early events (more weight given to early difference between treatments) and a value of q larger than 1 makes it more sensitive to late events.

An example of a treatment that may have a delayed effect on cognitive decline in dementia is EGb761®, a standardised extract produced from the dried leaves of the maidenhair tree (Ginkgo biloba), for which there is an indication of a beneficial effect on cognitive symptoms in elderly subjects with and without dementia (8., 9., 10., 11., 12., 13.). A study of the elderly women included in the EPIDOS registry (14) suggested that the risk of development of AD after seven years was reduced in patients receiving EGb761® for at least two years. More recently, twenty-year follow-up data from the PAQUID cohort, a large community based cohort study of cognitive decline, also suggested that long-term exposure to EGb761® is associated with a slower rate of cognitive decline measured with the MMSE (15).

We have recently completed a large, five-year randomised, placebo-controlled secondary prevention trial (GuidAge) in 2854 community-based subjects with spontaneous memory complaints to their family physicians to evaluate whether treatment with EGb761® can modify the risk of developing AD (16). This was a multicentre, double blind, randomised, parallel group, placebo-controlled study performed in a community setting in France between 2002 and 2009. The trial was registered on ClinicalTrials.gov (identifier: NCT00276510) and the methodology of the study has been described in detail elsewhere (17). The study enrolled 2 854 community-dwelling subjects aged >70 years without dementia at inclusion, who spontaneously consulted their family physician for a memory complaint. Subjects were randomised to treatment with EGb761® 120 mg bid or matching placebo for a period of up to 5 years. Participants underwent annual cognitive, functional and dementia evaluations at which diagnostic criteria for AD were assessed systematically. The primary efficacy endpoint was the time to conversion from memory complaint to Alzheimer’s type dementia, assessed on the basis of the DSM-IV (18) and NINCDS-ADRDA (19) diagnostic criteria.

During the study, 134 subjects developed AD over five years of treatment, 61/1406 in the EGb761® group and 73/1414 in the placebo group. According to the preplanned logrank test, there was no significant difference in the hazard function of conversion between the two treatment groups (p = 0.3058) (16). It should be noted that the conversion rate observed in the study was unexpectedly low, raising the question as to whether the GuidAge study should be considered an inconclusive study or a negative one. The a priori power calculations of the study were based on an anticipated total of 338 events over five years with a hazard ratio that was constant over time. This would have provided a power of 80% to detect a 25% difference in conversion rate between the placebo and EGb761® treatment arms using the logrank test. In fact, only 134 events occurred, reducing the actual power of the study to 42%. However, in a protocol-specified subgroup analysis in which conversion rates were compared for each year of the study, a significant treatment-by-time interaction for the incidence of AD was observed (p = 0.043), suggesting that the hazard ratio was not constant over time (Figure 1). Interestingly, the evolution over time of the hazard ratio for the treatment effect observed in the GuidAge study is similar to that previously reported in the EPIDOS study (14).

Figure 1.

Rates of conversion to dementia in the GuidAge study as a function of treatment duration. Hazard ratios are presented with their 95% confidence intervals. Grey columns: placebo group; black columns: EGb761® group

Given the significant interaction and similarities of evolution across studies it was hypothesized that if EGb761 has an effect it is a late one. To verify this working hypothesis, a post hoc Fleming-Harrington test was performed on the entire set of data. For this analysis, the value of the q parameter was set a priori to three and that of the p parameter to 0. Applying the Fleming-Harrington test, the hazard function for conversion to dementia in the placebo group was significantly different from that in the EGb761® treatment group (p = 0.0054), suggesting a late effect of EGb761®. A similar significant difference was observed in the per protocol population (p = 0.0137), as well as in sensitivity analyses in which the event definition was enlarged to include retrieved drop outs (p = 0.0029) or conversion to mixed dementia (p = 0.0012). Moreover, significant between-group differences were also observed using other parameterisations of the Fleming-Harrington test, where q is set to other values >1 and p maintained at 0 (Table 1).

Table 1.

Influence of the value of the q parameter on the p value of the Fleming Harrington test in the GuidAge study. The larger the value of q the larger the weight assigned to late events

p and q parameters	0 and 0 (Logrank)	0 and 1	0 and 2	0 and 3	0 and 4	0 and 5
Statistic	1.030	1.964	2.562	2.814	2.882	2.858
p value	0.304	0.049	0.010	0.004	0.003	0.002

Since this was a post hoc analysis, no definitive conclusions can be drawn as to the effectiveness of the treatment, but the findings do at least challenge the null hypothesis of no difference between treatment groups and provide a signal for a delayed effect of EGb761® on cognitive decline. This signal would need to be confirmed in another prospective study specifically testing the hypothesis of a late treatment effect using an appropriate statistical test.

Effective prevention strategies for AD, which is the most frequent neurodegenerative disease affecting the elderly, currently affecting 35 million people worldwide (20), would represent a powerful public health strategy for reducing the burden of disease (21). However, to date, no prevention strategy has been unequivocally demonstrated to be successful (22). Clinical trials to assess such strategies are costly and time-consuming, and for this reason, appropriate design, including the determination of the sample size and the statistical analysis to be employed, is critical.

In clinical trials using time-to-event analysis, there is generally no prior information as to how the hazard ratio for the event of interest may be expected to evolve over time. However, the statistical testing needs to be specified before the start of the trial. The Fleming-Harrington test allows appropriate testing of different patterns of hazard, but the appropriate parameterisation of the test (setting the values of p and q) will depend upon the expected evolution of the hazard ratio over time (high p parameter and low q value for an early effect and the opposite for a late effect). We have recently proposed a two-step strategy for testing treatment effects in which the best matching value of q can first be determined and the appropriate Fleming-Harrington test then implemented (23, 24). This strategy does not entail an unreasonable inflation of the sample size.

In conclusion, in clinical trials using time-to-event analysis in neurodegenerative diseases, the hazard ratio may evolve over time and this needs to be taken into account in the design of the study. This is illustrated by a post hoc analysis of data from the GuidAge study that provides a signal of the efficacy of long-term treatment with EGb761® in delaying the onset of AD. Potential late effects should be taken into consideration in the planned duration and sample size of future prevention studies and in the choice of statistical approach.

Conflict of Interest Disclosures

BS reports personal fees from Ipsen, during the conduct of the study and personal fees from Ipsen, outside the submitted work. SA has received consultancy fees, speaker’s fees or research funding from Beaufour-Ipsen, Eisai Inc., Eli Lilly, Lundbeck A/S, Nestle, Novartis, Pierre Fabre, and Servier. JFD reports grants from Ipsen, during the conduct of the study and grants from Ipsen, outside the submitted work. BD reports personal fees from Eli Lilly, grants from Pfizer, outside the submitted work. PR has received consultancy fees from Lundbeck and Ipsen. PJO, GB, FP, FP and JT received consultancy fees from Ipsen Pharma for their participation in the Guidage study’s Scientific Committee. PG and HMF are employees of Ipsen Pharma who funded the GuidAge Study. BV has received research grants or consultancy fees from Lilly, Lundbeck, Medivation, MSD, Nestle, Roche, Sanofi, Servier, Otsuka, Pierre Fabre, Regeneron, Biogen, and Taurx Therapeutics.

Ethical standards

Each subject was informed (both verbally and by receipt of written information) of the aims, methods, anticipated benefits, potential hazards and the discomfort the study may entail, and their right to abstain from participating in the study and to withdraw their consent at any time without affecting their medical care. Written consent was obtained at the time of screening for the study and before to perform all additional tests planned in the protocol amendments. Consent forms were personally signed and dated, by both the subject and the investigator, as a confirmation of consent. Subject information sheets, informed consent forms and protocol were approved by the Independent Ethics Committee (IEC) of Toulouse, France (CCPPRB Toulouse I - Comité Consultatif de Protection des Personnes se Prêtant à des Recherche Biomédicales de Toulouse I).