determining the minimum clinically important differences for outcomes in the domino trial

Determining the minimum clinically important differencesfor outcomes in the DOMINO trial

Robert Howard1, Patrick Phillips2, Tony Johnson2,3, John O’Brien4, Bart Sheehan5, James Lindesay6,Peter Bentham7, Alistair Burns8, Clive Ballard9, Clive Holmes10, Ian McKeith4, Robert Barber4, Tom Dening11,Craig Ritchie12, Rob Jones13, Ashley Baldwin14, Peter Passmore15, David Findlay16, Alan Hughes17,Ajay Macharouthu17,18, Sube Banerjee19, Roy Jones20, Martin Knapp21, Richard G. Brown22, Robin Jacoby23,Jessica Adams1, Mary Griffin1 and Richard Gray24

1King’s College London, Institute of Psychiatry, Department of Old Age Psychiatry, London, UK2MRC Clinical Trials Unit, London, UK3MRC Biostatistics Unit, Institute of Public Health, Cambridge University, Cambridge, UK4Institute for Ageing and Health, Campus for Ageing and Vitality, Newcastle University, UK5Health Sciences Research Institute, Warwick Medical School, University of Warwick, Coventry, UK6Department of Health Sciences, University of Leicester, Leicester General Hospital, Leicester, UK7Queen Elizabeth Psychiatric Hospital, Edgbaston, Birmingham, UK8School of Psychiatry and Behavioural Sciences, Wythenshawe Hospital, Manchester, UK9King’s College London, Wolfson Centre for Age Related Disease, London, UK10MARC, University of Southhampton, Moorgreen Hospital, Southampton, UK11Older People’s Mental Health Service, Fulbourn Hospital, Cambridge, UK12Centre for Mental Health, Claybrook Centre, Faculty of Medicine, Imperial College London, UK13RICE (the Research Institute for the Care of Older People), Royal United Hospital, Bath, UK145 Boroughs Partnership NHS Foundation Trust, UK15Ageing Group, Centre for Public Health, Queens University Belfast, Whitla Medical Building, Belfast, UK16Dundee Community Health Partnership, Royal Dundee Liff Hospital, Dundee, UK17Department of Geriatric Psychiatry, Inverclyde Royal Hospital, Inverclyde, UK18Mental Health Directorate, Crichton Royal Hospital, Dumfries, UK19King’s College London, Institute of Psychiatry, Section of Mental Health and Ageing, London, UK20Research Institute for Care of the Elderly, St Martin’s Hospital, Bath, UK21King’s College London, Institute of Psychiatry, Department of Economics of Mental Health, London, UK22King’s College London, Institute of Psychiatry, Department of Psychology, London, UK23Department of Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK24University of Birmingham Clinical Trials Unit, Edgbaston, UKCorrespondence to: Prof. R. Howard, E-mail: [email protected]

Background: Although less likely to be reported in clinical trials than expressions of the statisticalsignificance of differences in outcomes, whether or not a treatment has delivered a specified minimumclinically important difference (MCID) is also relevant to patients and their caregivers and doctors. Manydementia treatment randomised controlled trials (RCTs) have not reportedMCIDs and, where they havebeen done, observed differences have not reached these.

Methods:As part of the development of the Statistical Analysis Plan for the DOMINO trial, investigatorsmet to consider expert opinion- and distribution-based values for the MCID and triangulated these toprovide appropriate values for three outcome measures, the Standardised Mini-mental State Examin-ation (sMMSE), Bristol Activities of Daily Living Scale (BADLS) and Neuropsychiatric Inventory (NPI).Only standard deviations (SD) were presented to investigators who remained blind to treatmentallocation.

Results:Adoption of values for MCIDs based upon 0.4 of the SD of the change in score from baseline onthe sMMSE, BADLS and NPI in the first 127 participants to complete DOMINO yielded MCIDs of 1.4points for sMMSE, 3.5 for BADLS and 8.0 for NPI.

Conclusions:Reference to MCIDs is important for the full interpretation of the results of dementia trialsand those conducting such trials should be open about the way in which they have determined andchosen their values for the MCIDs. Copyright # 2010 John Wiley & Sons, Ltd.

RESEARCH ARTICLE

Copyright # 2010 John Wiley & Sons, Ltd. Int J Geriatr Psychiatry 2011; 26: 812–817.

Key words: dementia trials; analysis of results; clinically important differencesHistory: Received 6 July 2010; Accepted 6 July 2010; Published online16 September 2010 in Wiley Online Library(wileyonlinelibrary.com).DOI: 10.1002/gps.2607

Introduction

Clinical trials conventionally report their results interms of the differences between active and placebotreatments on a measured outcome together with its95% confidence interval (95% CI) and statisticalsignificance (p-value). Because neither the observeddifference nor its statistical significance may indicateclinical significance, the minimum clinically importantdifference (MCID) has been suggested as a more usefulmeasure of effectiveness. The MCID has been definedas ‘the smallest difference that patients perceive asbeneficial and which would mandate, in the absence oftroublesome side effects and excessive cost, a change inthe patient’s management’ (Jaeschke et al., 1989quoted in Burback et al., 1999). Published studieshave recommended the use of both distribution-based(for example standard deviation (SD) of distribution ofchange scores from baseline to outcome), and anchor-based (for example change defined by a set number ofpoints on a scale determined by clinical expertise, orachievement of a nominated target scale value againdetermined by clinical experience) methods todetermine MCID (Revicki et al., 2008; Stargardtet al., 2009; Barber et al., 2009). Results generated bythese methods can be integrated systematically usingthe method of triangulation (Leidy, 2006), which isparticularly valuable when distribution-based andanchor-based methods generate widely differing valuesfor the MCID (Leidy and Wyrwich, 2005).One of the earliest distribution-based methods of

identification of a clinically important change to beadvocated was Cohen’s effect size, a dimensionlessquantity, often derived by dividing mean differencebetween treatment groups at outcome by baseline SD,although Cohen describes other measures as well(Cohen, 1969). Using this method to compare groupaverages, a small effect size is conventionally 0.2, amedium effect size 0.5 and a large effect size 0.8. Cohenanchors these by reference to examples from psychol-ogy including intelligence tests and grades in the UKCivil Service. Increasingly, fractions of the SD have alsobeen used to estimate the MCID since this offers asimpler and apparently more transparent method thanalternatives. An influential systematic review of studies

that had reported MCIDs for health-related quality oflife measures reported a remarkable consistency amongthem, with convergence upon a value of 0.5 SD(Norman et al., 2003). The authors suggested thatone reason for the convergence upon 0.5 SD for theMCID in all of these studies might be the apparentupper limit of human ability to make absolutediscriminations within a perceptual task, which isreached at around seven different categories (Miller,1956). Hence, we are only capable of accuratelyidentifying the relative loudness of sounds or saltinessof tastes until presented with up to around sevendifferent values. This limit of human discrimination (1part in 7) equates to 0.46 SD or an effect size of between0.36 and 0.63 (Norman et al., 2003). Norman et al.(2003) suggested that the value of 0.5 SD should beused as a default for defining important patient- andclinician-perceived change.The Donepezil and Memantine in Moderate to

Severe Alzheimer’s Disease (DOMINO) trial(ISRCTN49545035) investigates whether treatmentwith donepezil or memantine alone or in combinationis better than placebo at maintaining cognitivefunction and activities of daily living in patients withmoderate to severe AD (Mini-mental State Examin-ation (MMSE) 5–15) who have previously been takingdonepezil (Jones et al., 2009). DOMINO is apragmatic, double-blind, randomised, placebo-con-trolled (double dummy), parallel group, 2� 2 factorialclinical trial in which participants are randomised toreceive (i) donepezil 10mg and memantine placebo,(ii) donepezil placebo and memantine 20mg, (iii)donepezil 10mg and memantine 20mg or (iv)donepezil placebo and memantine placebo for12 months. The co-primary outcomes are cognitionmeasured with the Standardised Mini-mental StateExamination (sMMSE) and function measured withthe Bristol Activities of Daily Living Scale (BADLS) at12 months from randomisation. As part of the TrialManagement Group’s meeting to finalise the StatisticalAnalysis Plan for the trial in May 2010, the principalinvestigators (PIs) decided upon the size of the MCIDsfor each of the three trial outcomes that should be builtinto the Analysis Plan. This paper reports themethodology that underpinned the decisions that set


Determination of clinically important differences 813

the MCIDs for the sMMSE, BADLS and Neuropsy-chiatric Inventory (NPI) for DOMINO.

Methods

At ameeting of the PIs of the DOMINO trial held at theHeadquarters of the Medical Research Council (MRC)in London in May 2010, two potential approaches forcalculating the MCID were discussed: expert opinionanchor-based and a distribution-based method usingdata collected from the trial.

(1) Investigator opinion. The DOMINO trial clinicianPIs are senior old age psychiatrists and geriatri-cians, all of whom prescribe dementia drugs andare involved in the conduct of independent andindustry-sponsored randomised controlled trials(RCTs) in this area. Opinions of PIs were soughtthrough email discussions prior to the meeting andduring face-to-face discussion chaired by the ChiefInvestigator (RH) at the meeting.

(2) Distribution-based. The Trial Statisticians pre-sented the values of the SDs of baseline measuresand change from baseline at 52 weeks data collectedfrom the 127 participants who had already com-pleted all trial assessments. No point estimates ofoutcomes or changes from baseline, only SDs blindto treatment allocation, were presented. The SD ofbaseline measures was calculated as the SD of allbaseline scores for the 127 participants ignoring thetreatment allocation. Using this method to calcu-late the SD in change from baseline results in an SDthat is correlated with any treatment differences atthe week 52 visit. Instead, the SD of change frombaseline removing the treatment effect was calcu-lated as the root mean squared error resulting fromthe one-way analysis of variance comparing thechange in baseline across treatment groups. The

group examined the size of MCID that would begenerated by the use of 0.5, 0.4 and 0.33 SD of thechange from baseline scores on the sMMSE,BADLS and NPI for the 127 trial participantswho had completed all assessments at the end of52 weeks of DOMINO.

Results

Investigator opinion (anchor-based) approach

The group agreed that although it had limitedfunctional relevance, the sMMSE was the trial measurethat clinicians, including many members of the PIgroup, were most familiar with and which wouldtherefore give an MCID with most salience. The groupwas not able to agree on a single value of the change insMMSE as an MCID, although there was generalagreement that this should lie somewhere within arange of between 1.0 point and 2.0 scale points. Therewas also agreement that an MCID that was based uponthe equivalent of 6-months deterioration in sMMSEscore, as suggested by the Food and Drug Adminis-tration (1989), had some face validity.

Distribution-based approach

The group reviewed the SDs for both the baseline andthe change from baseline scores on the sMMSE,BADLS and NPI for the 127 trial participants who hadcompleted all assessments at the end of 52 weeks fromrandomisation. These are shown in Table 1. The groupconsidered which of the given values; 0.5, 0.4 or0.33 SD would be most appropriate to choose for theMCID. Following the convention of Norman et al.(2003), a default choice of 0.5 SD of the change frombaseline score would give anMCID of 1.7 points for thesMMSE.

Table 1 Standard deviations of baseline and change from baseline scores for BADLS, sMMSE and NPI in first 127 DOMINO trial completers

Variable SD 0.33�SD 0.4�SD 0.5�SD

Baseline sMMSE 4.2 1.4 1.7 2.1Change in sMMSE from baseline to week 52 3.4 1.1 1.4 1.7Baseline BADLS 10.1 3.3 4.0 5.0Change in BADLS from baseline to week 52 8.7 2.9 3.5 4.3Baseline NPI 19.9 6.6 8.0 10.0Change in NPI from baseline to week 52 20.1 6.6 8.0 10.0

All 127 patients had non-missing scores at baseline and at the week 52 visit. The sMMSE is a measure of cognitive function ranging from 30 (unimpaired) to 0

(impaired). The BADLS is an assessment of activities of daily living in dementia ranging from 0 (unimpaired) to 60 (impaired). The NPI measures a caregiver’s

assessment of the frequency and severity of behavioural and psychiatric symptoms in dementia ranging from 0 (no disturbance) to 144 (maximum disturbance).


814 R. Howard et al.

Triangulation

For the sMMSE, considered first because of itspresumed salience, the group agreed that 0.5 SD ofthe change in score from baseline yielded a figure(1.7 points) that was too high as it was too close to theupper end of the range of 1.0–2.0 points that hademerged from the opinion-based MCID derivation.However, use of 0.4 SD gave a value of 1.4 points whichwas close to the middle of the opinion-based range andwhich coincides with the mean 6-month decline in ADquoted by the authors of the AD2000 study (AD2000Collaborative Group, 2004) from a review of pre-viously published decline rates (Burback et al., 1999).Adoption of 0.4 SD of the change in score frombaseline as a definition would give MCIDs of 3.5 pointsfor the BADLS and 8.0 points for the NPI.

Discussion

Despite a large clinical trial base, the importance andcost-effectiveness of drug treatments for Alzheimer’sdisease (AD) have been the subject of debate withinfluential reviews concluding both that cholinesteraseinhibitors should be a standard of care and that thescientific basis for recommending their use is ques-tionable (Cummings, 2004; Kaduszkiewicz et al.,2005). The UK National Institute of Health andClinical Excellence (NICE) undertook a technologyappraisal of these drugs in 2006 and the process leadingto their final guidance, that cholinesterase inhibitorsshould only be available to patients with moderate AD,was complex and subject to challenge (O’Brien, 2006).Controversy has arisen in this area largely because ofthe reported modest clinical effects of AD treatments.For example, a systematic review of published RCTsfound that on the commonest cognitive outcomereported, the 70 point AD assessment scale-cognitivesubscale (ADAS-Cog), differences between placeboand drug treated patients ranged from 1.5 points to 3.9points (Kaduszkiewicz et al., 2005), and noted thatthese differences all fell below the 4 points suggested bythe US Food and Drug Administration (1989) as anMCID.In the light of the number of people affected by AD,

the relative expense of drugs licensed for its treatmentand their apparently modest effects, it is surprising justhow little has been published about the size of MCIDthat would justify their use. Burback et al. (1999)collected responses from 161 geriatricians and neurol-ogists to the question: ‘from your experience followingdemented patients, what are the smallest changes in the

Folstein MMSE scores that are compatible with anoticeable change in the patient’s overall condition?’The meanMCID suggested was 3.72 MMSE points andboth the median and modal values of responses were3.0 points. These authors noted that none of the 13trials of tacrine published at the time had shown resultsthat reached this MCID, with the two trials that hadshown statistically significant superiority of drug overplacebo on MMSE score indicating mean changes ofonly 2.5 points (Knapp et al., 1994) and 2.7 points(Eagger et al., 1991). A more recent systematic reviewof measures of clinical significance used in dementiadrug trials found that 26 of 57 (46%) published trialshad discussed the clinical significance of their resultsand of these 19 had used opinion-based estimatesbased on cognitive scales and 8 based upon globalchange measures (Molnar et al., 2009). The mostcommonly cited MCIDs were a 4-point change inADASCog (seven studies) or change on a global scale(eight studies). None of the 57 trials reviewed hademployed a distribution-based measure of MCID.The AD2000 Trial (AD2000 Collaborative Group,

2004) was unusual and caused controversy byreporting positive statistically significant differencesfor the secondary outcomes of cognition (MMSEmeandifference 0.8; 95% CI 0.5–1.2) and functionality(BADLS mean difference 1.0; 95% CI 0.5–1.6) butconcluding that donepezil was not cost effective sincethese differences were below minimally relevantthresholds. This group set the MCIDs for AD2000on previously published values. Hence, the FDA hadsuggested as early as 1989 that a reversal of the naturalhistory of cognitive decline over 6 months constituteda clinically important difference and 1.4 MMSE pointscorresponded to 0.5 of the mean annual rate of declinein AD (Burback et al., 1999). The functionality MCIDin AD2000 was set at 3.0 BADLS points based upon apublished sensitivity to change study (Byrne et al.,2000) and discussion with the researchers who haddeveloped the BADLS. The minimum change thatwould constitute a clinically significant behaviouralresponse in AD2000 was based on earlier definitions of4.0 points (Mega et al., 1999) and 8.0 points (Kauferet al., 1996) improvement on the NPI.Given the controversy about the cost-effectiveness of

cholinesterase inhibitors and memantine in thetreatment of people with dementia it is importantthat the results of clinical trials are reported with clearreference to MCIDs as well as to conventionalsummary statistics such as mean differences (95%CI) and statistical significance (p-values). It is alsoimportant that values adopted for MCIDs are valid andthat their method of determination is transparent and



defensible. Choice of values based upon 0.4 SD of thechange in score from baseline of our main outcomemeasures resulted in MCIDs that are smaller than if wehad used the conventional 0.5 SD (Norman et al.,2003) but which, for the sMMSE at least, was close tothe middle of the range of scores suggested by expertopinion. The value of 1.4 sMMSE points that wearrived at in this way is also exactly the same as thatused for the MCID in AD2000. The other DOMINOMCIDs, derived in this way (0.4 SD of change frombaseline score), of 3.5 BADLS points and 8.0 NPIpoints are also very close to those chosen for AD2000although MCIDs in that trial were determined entirelyby expert opinion.Annual rates of decline on cognitive, functional and

behavioural measures are variable in AD. A meta-analysis of studies, mostly of patients from secondarycare services and research centres, found an averagerate of change of 3.3 points per year on the MMSE(Han et al., 2000), although a comprehensive reviewfound that rates varied considerably between studies,ranging from 0.8 to 4.0 points per year (Behl et al.,2005). Factors contributing to this include the fact thatMMSE decline is not linear across all stages of thedisease, survival bias and heterogeneity betweenstudied populations of patients. Similar variabilityhas been described in rates of functional decline(Aguero-Torres et al., 1998). Hence, extrapolation ofmean annual rates of decline on DOMINO outcomemeasures from earlier trials and cohort studies maylead to projections that are inherently flawed. Oneoption might be to wait until all the DOMINO trialdata has been collected and then use 50% of the annualrate of decline observed in the placebo-treatedparticipants to define the MCID. We have opted fora predetermined MCID to inform data analysis and toavoid apparent circularity in our choice of values forthe MCID. An important confirmation of the validityof our predetermined MCID, however, will be whetherthe annual rate of decline in the placebo participants isaround 2.8 MMSE points. If it is much greater orsmaller than this, then we may have to repeat thetriangulation process to re-estimate the MCID whichmay need to be greater or smaller than the value wehave estimated and cited in this paper.It is important to bear in mind that the estimation of

MCID between two treatment groups in a clinical trialrelates to mean differences between groups ofparticipants and does not fully capture the possibilitythat some individuals might be experiencing greater orlesser benefits that might be of importance to them,their care-givers or physicians. An intention-to-treatanalysis in a frail elderly population (with high rates of

drop outs and deaths) has the potential to deliver meanbetween group differences which mask substantialnumbers of individuals who do achieve definedimprovements on outcomes. An analogous situationis seen in antidepressant trials where the judgement ofefficacy is usually based on the proportions ofindividuals reaching a pre-defined target, for examplea 50% reduction in score on a depression rating scale(e.g. Wilson et al., 2001), rather than absolutedifferences in mean scores between treatment groups.In commonwithmany trials of anti-dementia drugs,

DOMINO may find statistically significant differencesbut not MCIDs between active treatment and placebogroups. Will this mean that the results of DOMINO arenegative or will the trial be subject to alternativeinterpretations as has been the case with the data fromAD2000? As the cholinesterase inhibitors come offpatent in 2011 and cheaper generic alternatives areavailable, much of the cost-effectiveness debatesurrounding their use will disappear. But these willalways be drugs with a potentially serious side-effectprofile (e.g. Gill et al., 2009) and decisions to advocatetheir use should be based upon clear evidence of thelikelihood of significant improvement that outweighssuch risks. We believe that deciding on MCIDs inadvance of knowledge of the results of trials representsgood practice in managing clinical trials and will aidsubsequent interpretation of results.

Conflicts of interests

None declared.

References

AD2000 Collaborative Group. 2004. Long-term donepezil treatment in 565 patientswith Alzheimer’s disease (AD2000): randomised double-blind trial. Lancet 363:2105–2115.

Aguero-Torres H, Fratiglioni L, Winblad B. 1998. Natural history of Alzheimer’sdisease and other dementia: review of the literature in the light of the findings fromthe Kungsholmen Project. Int J Geriatr Psychiatry 13: 755–766.

Key Points

� Calculation of minimum clinically importantdifferences helps the consumers of trial data todecide whether treatments are effective.

� Methods of estimation for the DOMINO trialconverged on values that were 0.4 SD of change inscore from baseline.

� MCIDs for DOMINO will be 1.4 points on thesMMSE, 3.5 points on the BADLS and 8.0 pointson the NPI.


816 R. Howard et al.

Barber MD, Spino C, Janz NK, et al. for the Pelvic Floor Disorder Network 2009. Theminimum important differences for the urinary scales of the pelvic floor distressinventory and pelvic floor impact questionnaire. Am J Obstet Gynaecol 200: 580e1–580e7.

Behl P, Stefurak TL, Black SE. 2005. Progress in clinical neurosciences: cognitivemarkers of progression in Alzheimer’s disease. Can J Neurol Sci 32: 140–151.

Burback D, Molnar FJ, St John P, Man-Son-Hing M. 1999. Key methodologicalfeatures of randomized controlled trials of Alzheimer’s disease therapy: minimalclinical difference, sample size and trial duration. Dement Geriatr Cogn Disord 10:534–540.

Byrne LMT,Wilson PMA, Bucks RS, Hughes AO,Wilcock GK. 2000. The sensitivity ofchange over time of the Bristol activities of daily living scale in Alzheimer’s disease.Int J Geriatr Psychiatry 15: 656–661.

Cohen J. 1969. Statistical Power Analysis for the Behavioural Sciences. Academic Press:London.

Cummings JL. 2004. Alzheimer’s disease. N Engl J Med 351: 56–67.Eagger SA, Levy R, Sahakian BJ. 1991. Tacrine in Alzheimer’s disease. Lancet 337: 989–

992.Food and Drug Administration. 1989. Peripheral and Central Nervous System Drugs

Advisory Committee Meeting, 7 July 1989, Rockville, MD. 227.Gill SS, Anderson GM, Fisher HD. 2009. Syncope and its consequences in patients with

dementia receiving cholinesterase inhibitors. Arch Intern Med 169: 867–873.Han L, Cole M, Bellavance F, McCusker J, Primeau F. 2000. Tracking cognitive decline

in Alzheimer’s disease using the mini-mental state examination: a meta-analysis. IntPsychogeriatr 12: 231–247.

Jaeschke R, Singer J, Guyatt GH. 1989. Measurement of health status: ascertaining theminimal clinically important difference. Control Clin Trials 10: 407–415.

Jones R, Sheehan B, Phillips P, et al. 2009. DOMINO-AD protocol: donepezil andmemantine in moderate to severe Alzheimer’s disease. Trials 10: 57.

Kaduszkiewicz H, Zimmerman T, Beck-Bornholdt HP, van den Bussche H. 2005.Cholinesterase inhibitors for patients with Alzheimer’s disease: systematic review ofrandomised clinical trials. Br Med J 331: 321–327.

Kaufer D, Cummings JL, Christine D. 1996. Effect of tacrine on behavioural symptomsin Alzheimer’s disease: an open-label study. J Geriatr Psychiatry Neurol 9: 1–6.

Knapp MJ, Knopman DS, Solomon PR. 1994. A 30-week randomized controlled trialof high-dose tacrine in patients with Alzheimer’s disease. J AmMed Assoc 271: 985–991.

Leidy NK. 2006. Evolving concepts in the measurement of treatment effects. Proc AmThorac Soc 3: 212–217.

Leidy NK, Wyrwich KW. 2005. Bridging the gap: using triangulation methodology toestimate minimal clinically important differences. J Chron Obstruct Pulmon Dis 2:1–5.

Mega M, Masterman DM, O’Connor SM, Barclay TR, Cummings JL. 1999. Thespectrum of behavioural responses to cholinesterase inhibitor therapy in Alzhei-mer’s disease. Arch Neurol 56: 1388–1393.

Miller GA. 1956. The magic number seven plus or minus two: some limits on ourcapacity for processing information. Psychol Rev 63: 81–97.

Molnar FJ, Man-Son-Hing M, Fergusson D. 2009. Systematic review of measures ofclinical significance employed in randomized controlled trials of drugs for demen-tia. J Am Geriatr Soc 57: 536–546.

Norman GR, Sloan JA,Wyrwich KW. 2003. Interpretation of changes in health-relatedquality of life: the remarkable universality of half a standard deviation.Med Care 41:582–592.

O’Brien JT. 2006. NICE and anti-dementia drugs: a triumph of health economics overclinical wisdom? Lancet Neurol 5: 994–996.

Revicki DA, Hays RD, Cella D, Sloan JA. 2008. Recommended methods for determin-ing responsiveness and minimally important differences for patient-reported out-comes. J Clin Epidemiol 61: 102–109.

Stargardt T, Gonder-Frederick L, Krobot KJ, Alexander CM. 2009. Fear of hypogly-caemia: defining a minimum clinically important difference in patients with type 2diabetes. Health Qual Life Outcomes 7: 91.

Wilson K, Mottram PG, Sivanantham A, Nightingale A. 2001. Antidepressants versusplacebo for the depressed elderly. Cochrane Database Syst Rev Issue 1 Art No.CD000561.



determining the minimum clinically important differences for outcomes in the domino trial

Documents