Skip Navigation



Age and Ageing Advance Access published online on December 23, 2008
Age and Ageing, doi:10.1093/ageing/afn279
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Supplementary Data
Right arrow All Versions of this Article:
38/2/151    most recent
afn279v1
Right arrow E-Letters: Submit a response to the article
Right arrow E-letters: View responses
Right arrow Alert me when this article is cited
Right arrow Alert me when E-letters are posted
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Donaldson, M. G.
Right arrow Articles by Khan, K. M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Donaldson, M. G.
Right arrow Articles by Khan, K. M.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

© The Author 2008. Published by Oxford University Press on behalf of the British Geriatrics Society. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org

Systematic Review

Analysis of recurrent events: a systematic review of randomised controlled trials of interventions to prevent falls

Meghan G. Donaldson1, Boris Sobolev2,3, Wendy L. Cook4, Patti A. Janssen2 and Karim M. Khan5

1 San Francisco Coordinating Centre, California Pacific Medical Center Research Institute, San Francisco, CA, USA
2 Department of Health Care and Epidemiology, Faculty of Medicine, University of British Columbia, Vancouver, British Columbia, Canada
3 Centre for Clinical Epidemiology and Evaluation, Vancouver Coastal Health Research Institute, University of British Columbia, Vancouver, British Columbia, Canada
4 Division of Geriatric Medicine, Faculty of Medicine, University of British Columbia, Vancouver, British Columbia, Canada
5 Centre for Hip Health and Mobility, Vancouver Coastal Health Research Institute and University of British Columbia, Vancouver, British Columbia, Canada

Address correspondence to: M. G. Donaldson, San Francisco Coordinating Centre, 185 Berry Street, Lobby 4, Suite 5700, San Francisco, CA 94107, USA. Tel: (+1) 415 600 7427; Fax: (+1) 415 514 8150. Email: mdonaldson{at}sfcc-cpmc.net


   Abstract
Top
 Abstract
Introduction
 Methods
 Results
 Discussion
 Key points
 Supplementary data
 References
 
Rationale: there are several well-developed statistical methods for analysing recurrent events. Although there are guidelines for reporting the design and methodology of randomised controlled trials (RCTs), analysis guidelines do not exist to guide the analysis for RCTs with recurrent events. Application of statistical methods that do not account for recurrent events may provide erroneous results when used to test the efficacy of an intervention. It is unknown what proportion of RCTs of falls prevention studies have utilised statistical methods that incorporate recurrent events.

Methods: we conducted a systematic review of RCTs of interventions to prevent falls in community-dwelling older persons. We searched Medline from 1994 to November 2006. We determined the proportion of studies that reported using three statistical methods appropriate for the analysis of recurrent events (negative binomial regression, Andersen–Gill extension of the Cox model and the WLW marginal model).

Results: fewer than one-third of 83 papers that reported falls as an outcome utilised any appropriate statistical method (negative binomial regression, Andersen–Gill extension of the Cox model and Cox marginal model) to analyse recurrent events and fewer than 15% utilised graphical methods to represent falls data.

Conclusion: RCTs that have a recurrent event end-point should include an analysis appropriate for recurrent event data such as negative binomial regression, Andersen–Gill extension of the Cox model and/or the WLW marginal model. We recommend that researchers and clinicians seek consultation with a statistician with expertise in recurrent event methodology.

Keywords: recurrent events, falls, elderly


   Introduction
Top
 Abstract
 Introduction
Methods
 Results
 Discussion
 Key points
 Supplementary data
 References
 
Evidence-based medicine has been promoted by the use of standards for reporting the design and methodology of randomised controlled trials (RCTs). An ever-increasing number of journals require that authors adhere to the Consolidated Standards for Reporting Trials (CONSORT) statement when reporting clinical trials [1]. The aim of that statement is to improve the consistency of data presentation so that readers can better assess the strengths and limitations of an RCT. To date, there are no guidelines for reporting events where the outcome can occur recurrently (e.g. falls, fractures, certain cancers and infections, chronic disease exacerbations).

To date, RCTs of falls prevention strategies most commonly report the time to first event and the proportion of participants who experience a fall. If only the time to first event is reported, as in Cox proportional hazard survival analysis, the study may conclude that the treatment is ineffective when the intervention may influence recurrence of the events 2, 3 and so on (see [2] for an example). Comparing only the proportion of participants who experience a fall within a time period is analogous to a time to first event analysis. Also, comparing proportion of participants who experience a fall discards valuable prospectively collected falls data in participants with unequal follow-up. The Prevention of Falls Network Europe (ProFaNE) has developed a common outcome data set for fall injury prevention trials [3]. These guidelines and a recent systematic review of definitions and methods of measuring falls in RCTs [4] suggest that trials report the rate of falls. However, at present there are no papers outlining specific statistical methods for analysing recurrent events such as falls.

Because of the lack of a definitive paper on how to analyse and report recurrent events such as falls, we systematically reviewed the statistical methods used in falls prevention RCTs to determine whether the proportion of RCTs using appropriate methods has increased from the period 1994–99 to 2000–06.


   Methods
Top
 Abstract
 Introduction
 Methods
Results
 Discussion
 Key points
 Supplementary data
 References
 
We searched Medline for articles published between 1994 (the first successful RCT to prevent falls in community-dwelling older adults was published [5]) and November 2006 inclusive. The search query consisted of a Boolean combination of relevant text words and database subject headings describing falls prevention in community-dwelling older persons (Table 1). The elements of the search query used to identify RCTs have been previously published [6]. We included all English language articles published in peer-reviewed journals that were RCTs in community-dwelling persons aged 60 years and older that reported falls as an outcome. We included articles that reported individual level meta-analyses. We also included articles that reported secondary analyses of a previously published trial. No assessment of quality was undertaken for each included trial.


View this table:
[in this window]
[in a new window]

  		Table 1. Search strategy and number of papers retained at each stage

 
The search strategy was validated against 10 authoritative articles in the field which were identified a priori [5, 7–15]. Included articles were also compared to references from two recent systematic reviews of falls prevention to assess sensitivity of our search strategy [16, 17].

We screened the title and abstract of the selected articles for the inclusion criteria and the full articles were retrieved if relevant. These articles were assessed independently by two of the authors (M.G.D. and B.S.).

We developed a data abstraction form based on previous literature to standardise the description of methods for each study [18–20]. It included study duration, number of dropouts due to death, duration of follow-up time, method of falls ascertainment, graphs and statistical methods with graphical display (23) and inferential statistics (3).

The reviewers independently read each article and completed the checklist related to statistical methods (19 items) (M.G.D. and B.S.). The checklists for both reviewers were compared for each article and discrepancies were resolved by consensus.

We report the prevalence of statistical methods using recurrent events by proportion of all articles. We compared the prevalence of these methods between two publication intervals, 1994–99 and 2000–06 using the chi-square statistic.


   Results
Top
 Abstract
 Introduction
 Methods
 Results
Discussion
 Key points
 Supplementary data
 References
 
The flow of trials through the systematic review is displayed (Figure 1). Of the 83 articles included, 77 (93%) were identified by the Medline search and the remaining articles were identified by the reference lists from two recent systematic reviews. Among the included articles, two were individual-level meta-analyses (see references 61 and 63); references 57 and 71 report results from the same trial and reference 10 is a 2-year follow-up of reference 11. Reference 64 is a re-analysis of references 11 and 65. There were six cluster RCTs (references 16, 39, 44, 71, 74 and 83; see Appendix 1 available on Age and Ageing online for the references cited in this paragraph).


Figure 1
View larger version (16K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 1. Assessment of studies for inclusion in RCTs of recurrent events.

 
The proportion of articles in the two publication intervals, 1994–99 and 2000–06, was 28% (23/83) and 72% (60/83), respectively. The studies included were published predominately in geriatrics and ageing journals (54%, 45/83) and general medical journals (22%, 18/83). The studies included between 18 and 5,292 participants (mean 504, median 236) and reported an observation period between 5 and 260 weeks (mean 55, median 52). Nearly all studies (92%, 76/83) reported variable participant follow-up times and lost between 0 and 17% of participants due to death. The methods for ascertaining falls in the 83 studies were monthly diary (33%, 27/83), recall at the final follow-up (34%, 28/83), weekly diary (7%, 6/83), monthly telephone call by the investigator (6%, 5/83) and weekly telephone call by the investigator (1%, 1/83). The method of ascertaining falls was not reported in 6 of the 83 studies.

All 10 of the ‘key’ articles identified a priori by the investigators were retrieved by the search. The rate of agreement on analysis method was 91% for specified statistical methods. Discrepancies were resolved by consensus.

The prevalence of the statistical methods with a graphical display by year of publication is shown in Table 2. The most prevalent statistical method with a graphical display was the Kaplan–Meier method (8/83) where authors estimated the cumulative probability of surviving and remaining free of a fall over the follow-up time. None of the articles included in this study used the mean cumulative function to display the average number of falls per participant by a certain time. We did not observe an increase in the proportion of articles utilising graphical displays to depict recurrent events from 1994–99 to 2000–06.


View this table:
[in this window]
[in a new window]

  		Table 2. Statistical methods with graphical display and number of papers (proportion) by publication year

 
Table 3 details the prevalence of the inferential statistical methods by year of publication. The most prevalent inferential statistical method reported was proportion of participants having at least one fall (47/83, 57%). Less than one-third of the articles reported at least one of the three recurrent event statistical methods (26/83). The prevalence of the negative binomial regression model, Andersen–Gill model and WLW marginal model was 25, 7 and 4%, respectively. There was no significant increase in the proportion of articles reporting negative binomial regression and marginal models. The proportion of articles reporting the Andersen–Gill model was significantly less in the period 2000–06 compared to 1994–99 ({chi}2, P = 0.03).


View this table:
[in this window]
[in a new window]

  		Table 3. Inferential statistics and number of papers (proportion) by publication year

 

   Discussion
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
Key points
 Supplementary data
 References
 
These findings suggest that in falls-related RCTs, the three well-established statistical methods for repeated events—negative binomial regression model, Andersen–Gill model and WLW marginal model—are under-used. As these methods have not been used consistently, the validity of these studies diminishes. Although the purpose of this systematic review was not to directly compare the results from these three methods using either simulation studies or existing data, the comparison of these methods and other methods for recurrent events has been published elsewhere [18, 21].

Recurrent events are the primary outcomes in many clinical scenarios and not just falls. They are relevant to many conditions of aging such as hospitalisations, fractures and upper respiratory tract infections. There has been a considerable debate as to analysis methods for recurrent events. On the one hand, it was suggested that simply reporting the events per person-year and ignoring the dependence of multiple events within participants is seriously flawed and investigators should report only the first event, using both proportion and time to event analysis [22]. On the other hand, authors have argued for reporting the rate of recurrent events to avoid incorrect conclusions by ignoring valuable information such as when the intervention fails to impact the first event but nevertheless influences subsequent events [2]. This is illustrated in an RCT of expedited cataract surgery to prevent falls conducted by Harwood and colleagues [23]. There was no significant difference between groups using time to first fall; however, when the rate of falls was compared, the rate of falls was 33% lower in the intervention group compared to the control group. Furthermore, sample size can potentially be reduced when all events are considered in the analysis compared with time to first event analysis [24]. However, these calculations are not straightforward and factors such as subject accrual, duration of follow-up, event rate parameters and dropout rates must be considered (please see Appendix 2 for further details, available on Age and Ageing online).

The mean cumulative function provides the average cumulative number of events per subject over time and the corresponding 95% confidence interval [25–27]. The MCF can be plotted, and the graphs allow for an easy interpretation of the average number of events expected in one participant at a certain time. It can also illustrate (i) when interventions begin to take effect (where the curves for the intervention and control group begin to diverge) and (ii) determine on average how many events are prevented per participant by a certain time (the difference between the two curves at any given time). Finally, the mean cumulative function graph may illustrate how intervention and control groups behave over time. For example, these graphs may help to determine when an intervention is flagging (for whatever reason) and may prompt a reassessment and initiate changes to the intervention program (e.g. compliance may be a problem). Please see Appendix 2 for more details and an example of the MCF curves (available on Age and Ageing online).

To our knowledge, this is the first systematic review in the falls literature to evaluate the prevalence of statistical methods. We searched Medline and used the references from two recent systematic reviews to obtain 83 articles; although we may have missed articles, it is likely that we missed both articles that did and did not utilise the three appropriate statistical methods. We recognise that the three methods discussed in detail do not cover all the options (e.g., semiparametric method developed by Lin and colleagues [28]), nor do we address other complex analysis issues that arise in the analysis of recurrent events. We direct the reader to other literature that addresses fine statistical details further [20, 24, 26, 29–31].

Although we contend that our findings have relevance to clinical conditions where outcomes occur recurrently, this study only included RCTs that reported falls. Notably, a recent systematic review of hospitalisations suggests that the appropriate statistical methods are also under-utilised [19]. Therefore, we recommend that clinicians and researchers consult with statisticians with expertise in the area of recurrent event analysis to help choose an appropriate statistical model that will answer the study question.


   Key points
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Key points
Supplementary data
 References
 

  • Falls is a major public health problem that occurs repeatedly in seniors.
  • There are several accepted and available statistical methods to analyse recurrent events such as falls including negative binomial regression, Andersen–Gill extension of the Cox model and marginal Cox regression models. Other methods such as time to first fall and proportion of fallers are analysis methods that can provide erroneous results when used to test the efficacy of an intervention that may affect subsequent events.
  • Our systematic review suggests that fewer than one-third of 83 papers that reported falls as an outcome utilised an appropriate statistical method to analyse recurrent events.
  • Our data suggest the need for guiding bodies such as CONSORT to include analysis standards for RCTS that have a recurrent event end-point.


   Conflicts of interest
 
There are no conflicts of interest.


   Supplementary data
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Key points
 Supplementary data
References
 
Supplementary data are available at Age and Ageing online.


   References
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Key points
 Supplementary data
 References
 

  1. Moher D, Schulz KF, Altman DG. The CONSORT statement: revised recommendations for improving the quality of reports of parallel-group randomised trials. Lancet (2001) 357:1191–4.[CrossRef][Web of Science][Medline]
  2. Glynn RJ, Buring JE. Ways of measuring rates of recurrent events. Br Med J (1996) 312:364–7.[Free Full Text]
  3. Lamb SE, Jorstad-Stein EC, Hauer K, Becker C. Development of a common outcome data set for fall injury prevention trials: the Prevention of Falls Network Europe consensus. J Am Geriatr Soc (2005) 53:1618–22.[CrossRef][Web of Science][Medline]
  4. Hauer K, Lamb SE, Jorstad EC, Todd C, Becker C. Systematic review of definitions and methods of measuring falls in randomised controlled fall prevention trials. Age Ageing (2006) 35:5–10.[Abstract/Free Full Text]
  5. Tinetti ME, Baker DI, McAvay G, et al. A multifactorial intervention to reduce the risk of falling among elderly people living in the community. N Engl J Med (1994) 331:821–7.[Abstract/Free Full Text]
  6. Haynes RB, McKibbon KA, Wilczynski NL, Walter SD, Werre SR. Optimal search strategies for retrieving scientifically strong studies of treatment from Medline: analytical survey. Br Med J (2005) 330:1179.[Abstract/Free Full Text]
  7. Campbell AJ, Roberton MC, Gardner MM, Norton RN, Buchner DM. Psychotropic medicine withdrawal and a home-based exercise program to prevent falls: a randomized controlled trial. J Am Geriatr Soc (1999) 47:850–3.[Web of Science][Medline]
  8. Campbell AJ, Robertson MC, Gardner MM, Norton RN, Tilyard MW, Buchner DM. Randomized controlled trial of a general practice programme of home based exercise to prevent falls in elderly women. Br Med J (1997) 315:1065–9.[Abstract/Free Full Text]
  9. Campbell AJ, Robertson MC, La Grow SJ, et al. Randomised controlled trial of prevention of falls in people aged > or = 75 with severe visual impairment: the VIP trial. Br Med J (2005) 331: 817.
  10. Robertson MC, Devlin N, Gardner MM, Campbell AJ. Effectiveness and economic evaluation of a nurse delivered home exercise programme to prevent falls: 1. Randomised controlled trial. In: Br Med J (2001) 322:697–701.[Abstract/Free Full Text]
  11. Robertson MC, Gardner MM, Devlin N, McGee R, Campbell AJ. Effectiveness and economic evaluation of a nurse delivered home exercise programme to prevent falls 2: controlled trial in multiple centres. In: Br Med J (2001) 322:701–4.[Abstract/Free Full Text]
  12. Close J, Ellis M, Hooper R, Gluckman E, Jackson S, Swift C. Prevention of falls in the elderly trial (PROFET): a randomized controlled trial. Lancet (1999) 353:93–7.[CrossRef][Web of Science][Medline]
  13. Hogan D, MacDonald F, Betts J, et al. A randomized controlled trail of a community-based consultation service to prevent falls. Can Med Assoc J (2001) 165:537–43.[Abstract/Free Full Text]
  14. Wolf SL, Barnhart HX, Kutner NG, McNeely E, Coogler C, Xu T. Reducing frailty and falls in older persons: an investigation of Tai Chi and computerized balance training. Atlanta FICSIT Group. Frailty and Injuries: Cooperative Studies of Intervention Techniques. J Am Geriatr Soc (1996) 44:489–97.[Web of Science][Medline]
  15. Lord SR, Ward JA, Williams P, Strudwick M. The effect of a 12-month exercise trial on balance, strength and falls in older women: a randomized controlled trial. J Am Geriatr Soc (1995) 43:1198–206.[Web of Science][Medline]
  16. Kannus P, Sievanen H, Palvanen M, Jarvinen T, Parkkari J. Prevention of falls and consequent injuries in elderly people. Lancet (2005) 366:1885–93.[CrossRef][Web of Science][Medline]
  17. Chang JT, Morton SC, Rubenstein LZ, et al. Interventions for the prevention of falls in older adults: systematic review and meta-analysis of randomised clinical trials. Br Med J (2004) 328:680.[Abstract/Free Full Text]
  18. Robertson MC, Campbell AJ, Herbison P. Statistical analysis of efficacy in falls prevention trials. J Gerontol A Biol Sci Med Sci (2005) 60:530–4.[Abstract/Free Full Text]
  19. Metcalfe C, Thompson SG, Cowie MR, Sharples LD. The use of hospital admission data as a measure of outcome in clinical studies of heart failure. Eur Heart J (2003) 24:105–12.[Abstract/Free Full Text]
  20. Cook RJ, Lawless JF. Analysis of repeated events. Stat Methods Med Res (2002) 11:141–66.[Abstract/Free Full Text]
  21. Kuramoto L, Sobolev BG, Donaldson MG. On reporting results from randomized controlled trials with recurrent events. BMC Med Res Methodol (2008) 8:35.[CrossRef][Medline]
  22. Windeler J, Lange S. Events per person year—a dubious concept. Br Med J (1995) 310:454–6.[Free Full Text]
  23. Harwood RH, Foss AJ, Osborn F, Gregson RM, Zaman A, Masud T. Falls and health status in elderly women following first eye cataract surgery: a randomised controlled trial. Br J Ophthalmol (2005) 89:53–9.[Abstract/Free Full Text]
  24. Cook RJ. The design and analysis of randomized trials with recurrent events. Stat Med (1995) 14:2081–98.[Web of Science][Medline]
  25. Nelson W. Confidence limits for recurrence data: applied to cost or number of repairs. Technometrics (1995) 37:147–57.[CrossRef][Web of Science]
  26. Nelson WB. Recurrent Events Data Analysis for Product Repairs, Disease Recurrences, and Other Applications (2003) Philadelphia, PA: SIAM.
  27. Doganaksoy N, Nelson W. A method to compare two samples of recurrence data. Lifetime Data Anal (1998) 4:51–63.[CrossRef][Web of Science][Medline]
  28. Lin DY, Wei LJ, Yang I, Ying Z. Semiparametric regression for the mean and rate functions of recurrent events. J R Statist Soc B (2000) 62:711–30.[CrossRef]
  29. Cook RJ, Lawless JF. The Statistical Analysis of Recurrent Events (2007) New York: Springer.
  30. Hosmer D, Lemeshow S, May S. Applied Survival Analysis: Regression Modelling of Time to Event Data (2008) Hoboken, NJ: Wiley.
  31. Wang MC, Qui J, Chiang CT. Analyzing recurrent event data with informative censoring. JASA (2001) 96:1057–65.
Received 31 July 2008; accepted in revised form 2 October 2008.


Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


E-letters:

Read all E-letters

Recurrent event analysis
Andreas Wienke
Age and Ageing, 26 Oct 2009 [Full text]
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Supplementary Data
Right arrow All Versions of this Article:
38/2/151    most recent
afn279v1
Right arrow E-Letters: Submit a response to the article
Right arrow E-letters: View responses
Right arrow Alert me when this article is cited
Right arrow Alert me when E-letters are posted
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Donaldson, M. G.
Right arrow Articles by Khan, K. M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Donaldson, M. G.
Right arrow Articles by Khan, K. M.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?