ANALYSIS CPD

The “1-year-death number needed to treat” for comparing the impact of distinct interventions on patient outcomes

Brett N. Hryciw MD, Finlay A. McAlister MD MSc, Meltem Tuna MSc PhD, Carl van Walraven MD MSc n Cite as: CMAJ 2019 November 11;191:E1242-9. doi: 10.1503/cmaj.190367

fter concluding that a randomized controlled trial (RCT) is both internally and externally valid, consumers of evi­ KEY POINTS dence may want to consider an intervention’s overall • Cross-study comparisons of health care interventions are importanceA compared with other interventions or health technol­ difficult because of varying patient populations, comparators, ogies. Such an exercise contributes to our perception of an inter­ concomitant treatments and outcomes, which all affect the vention’s importance to health. This process could also help health absolute risk of events. care organizations determine which of an assortment of new tech­ • The 1-year-death number needed to treat (NNT) standardizes nologies should be introduced first, or identify processes of care comparisons of treatment impact by focusing on all-cause mortality 1 year from treatment initiation using NNT that should be subjected to quality-of-care reviews and optimiza­ methodology to account for distinct baseline risks. tion. Finally, it might also help with treatment selection for an indi­ • It is interpreted as the annual number of people requiring vidual patient; for example, in a patient with newly diagnosed left treatment and observation to avoid 1 death, with most beneficial ventricular systolic dysfunction, should one start treatment with treatments having positive values closest to 0 and most harmful angiotensin-converting-enzyme (ACE) inhibition or β-blockade? treatments having negative values closest to 0. However, making comparisons across trials and between dif­ • The 1-year-death NNT permits the standardized comparison of ferent technologies can be difficult because studies differ with diverse health care technologies, which could allow physicians to respect to patient populations, comparators (placebo v. active prioritize distinct treatment options for an individual patient or health care organizations to determine which new technologies control interventions or active controls of differing efficacy), con­ to introduce first. comitant treatments, follow-up times and outcomes. Compari­ sons across trials from different eras are even more difficult given secular trends in health outcomes over time. Each of these fac­ tic summarizes the influence of treatment on patient survival by tors can influence the baseline absolute risk of an event and, accounting for between-study variations in baseline death risk therefore, interstudy comparisons. and follow-up duration. It is calculated as To help with this, we propose here a method to compare the impact of diverse interventions applied to different populations 1-year death NNT = outcome time/ARR over varying observation times and with distinct outcomes using a summary statistic we call the 1-year-death number needed to in which ARR is the absolute risk reduction (calculated as death treat (NNT). We illustrate the application of this method using a riskcontrol group – death riskintervention group, with death risks in each treat­ convenience sample of landmark RCTs. We then present the ment arm expressed as proportions from 0 to 1 and calculated as studied interventions in a league table of 1-year-death NNTs to the number dying divided by the number randomly assigned). show how the statistic can be used to gauge the impact of inter­ Outcome time is the number of years after randomization when ventions on patient outcomes. death status was measured in the study. If the study did not fol­ low all patients for the same amount of time, the outcome time What is the 1-year-death NNT? is the average observation time. Some interventions have a finite treatment duration, with death status measured later on; in such Although reported outcomes vary extensively between studies, cases, outcome time occurs at the end of the “treatment cycle,” all-cause mortality is commonly reported and is an outcome which includes both treatment time and the subsequent lag time whose clinical importance is arguably independent of both the before measurement of death status. patient population and treatment type. We therefore used all- The 1-year-death NNT can have values that range from nega­ cause mortality as the basis for the 1-year-death NNT. This statis­ tive to positive infinity. The most beneficial treatments will have

E1242 CMAJ | NOVEMBER 11, 2019 | VOLUME 191 | ISSUE 45 © 2019 Joule Inc. or its licensors ANALYSIS ­ and and E1243 21 = 4), estab 4), = n ). 30 = 3), cirrhosis and its complications and its complications = 3), cirrhosis = 13, 41.9%) involved finite treatment finite treatment = 13, 41.9%) involved = 5), intensive care issues ( issues care intensive 5), = = 2), and 1 each of diabetes, premature premature 1 each of diabetes, = 2), and n Death risks in control groups varied exten­ Death risks in control warfarin for symptomatic intracranial arter­ 10 18 Given this study’s treatment cycle duration of treatment cycle this study’s Given death NNT (for prolonged The largest 1-year-death NNT (for prolonged 2 ISSUE 45 ISSUE | 2–12 outcome time was 1.7 years after randomization 1.7 years after randomization outcome time was encainide or flecainide for frequent premature ventricular 13 hepatic portosystemic shunting in acute variceal bleeding, variceal bleeding, hepatic portosystemic shunting in acute death NNT (in absolute values) was 18. The median 1-year-death NNT (in absolute values) was 18. Four trials (12.9%) found a significantly increased death risk Four trials (12.9%) found a significantly increased death risk Five trials (16.1%) involved a procedural intervention, 13 trials intervention, involved a procedural Five trials (16.1%) = 3), arrhythmia (n = 3), arrhythmia n VOLUME 191 VOLUME sively between studies (range 2.8% to 88.5%, median 22.7%, (range 2.8% to 88.5%, median 22.7%, sively between studies of 27 studies reported on interventions that mean 26.7%). A total ­ and the remaining 4 identified interven decreased death risk, risk. tions that increased death (in of lives per year greatest number saving the The intervention premature during betamethasone was lives) neonatal case, this labour, with a 1-year-death NNT of 0.16 (95% confidence interval to 0.64). [CI] 0.09 death NNT –21, contractions after myocardial infarction [1-year-death NNT –21, 95% CI –47 to –13], tamoxifen after initial 5 years of treatment in early breast cancer) tamoxifen after initial 5 years of treatment was 556 (95% CI 272 to ∞). with the study’s intervention (including strict glucose control for critical care patients [1-year-death NNT –9.4, 95% CI –63.8 to –5.1], What can we learn from estimating the 1-year- death NNT? By calculating the 1-year-death NNT, we were able to standard­ ize both changes in death risk between treatment groups and the time after randomization when death risk was calculated. This let us directly compare the impact of very distinct health care tech­ nologies on survival (Table 2). These benefits ranged from a life ial stenosis [1-year-death NNT –34, 95% CI –148 to –19], intensive glucose control in high-risk patients with type 2 diabe­ tes [1-year-death NNT –334, 95% CI –∞ to –54] ­ less than 1 year, this means (Figure 1) that 0.16 mothers in pre every week for a mature labour required 1 day of betamethasone and observed) to year (for a total of 0.16 × 52 = 8.3 people treated beneficial and associ­ avoid 1 death. Eleven interventions were in ated with a 1-year-death NNT of less than 10 (betamethasone ventilation in severe adult respiratory premature labour, prone hepatitis, goal- distress syndrome, prednisolone in alcoholic bac­ spontaneous in albumin sepsis, in resuscitation directed of amphotericin B in terial peritonitis, voriconazole instead in dexamethasone aspergillosis, invasive immunocompromised transjugular resuscitation, after hypothermia meningitis, acute intra­ in severe congestive tranexamic acid in trauma, and enalapril heart failure). nary artery disease ( disease artery nary = disease (n cardiovascular at a high risk of attaining) lished (or diseases (n 3), infectious ( to 2013. dates ranged from 1972 cancer. Publication labour and treatments (i.e., daily medication), (41.9%) involved continuous (n and all other studies 6 hours to 3 months (Table 2). Studies durations ranging from mean of 5912 patients (range 61 to 45 852). randomly assigned a The mean Patient loss to follow-up was minimal in all (range 1 wk to 7.6 yr). studies except one. | per for h cycles This is a col­ 1 deat 1 required changing,” with a a with changing,” NOVEMBER 11, 2019 NOVEMBER sequentially avoid cycle (with | people 1 yr) to CMAJ repeated No. of treatment < 1 yr cycle duration yr and ≥ 1 Treatment avoid people to death treatment 1 observation Annual no. of requiring summarizes the 31 RCTs reporting a significant differ­ significant a reporting RCTs the 31 summarizes 1 Table

How can the 1-year-death NNT be applied? How can the 1-year-death NNT To illustrate application of the 1-year-death NNT, we used as our sampling frame all RCTs included in 2 Minute Medicine’s The Clas- sics in Medicine: Summaries of the Landmark Trials. positive values closest to zero. Negative 1-year-death NNT val­ positive values closest to zero. Negative death risk (with the ues indicate interventions that increase the NNT being interpreted as 1-year-death absolute value of the the 1-year- Interpretation of harm”). “number needed to treat to duration (Figure 1).death NNT depends on the treatment cycle the 1-year-death NNT is If the treatment cycle is 1 year or longer, and observa­ the annual number of people requiring treatment is less than a year, tion to avoid 1 death. If the treatment cycle per the 1-year-death NNT is the number of people required for 1 year) to treatment cycle (with cycles repeated sequentially -year-death NNT of 5. If avoid 1 death. For example, consider a 1 that 5 people need the treatment cycle is 2 years, this means avoid a single annual treatment and observation for 1 year to this means that 5 peo­ death; if the treatment cycle is 6 months, 6 months twice (for a ple need treatment and observation for a single to avoid -years) patient or 5 people 10 treated of total annual death. lation of 190 “landmark clinical studies,” defined by the authors as those identified by academic clinicians in general internal - “practice as subspecialties its and medicine focus on common medical conditions (Dr. Andrew Cheung, St. Joseph’s Healthcare Hamilton, Hamilton, Ont.: personal com­ munication, 2019). We limited our convenience sample to RCTs of treatment evaluations least biased the these provide (since effects) and those reporting all-cause mortality (regardless of methods Further outcome). primary study’s the was it whether and results are presented in Appendix 1 (available at www.cmaj. ca/lookup/suppl/doi:10.1503/cmaj.190367/-/DC1). ence in all-cause death risk between treatment groups. These trials included patients with congestive heart failure (n = 8), coro­ death number needed to treat. to treat. 1-year-death number needed Figure 1: Interpretation of the of treatment to mea­ Note: treatment cycle duration = time from start surement of death status. ANALYSIS E1244 Pitt et al. Sort et al. Rose et al. van deBeek de Gansand Study mortality Table 1(part of 1of landmarkrandomized 2):Summary controlled trialsinwhichtreatments significantly reduced all-cause Echt et al. Finfer et al. Group Trial Study CONSENSUS Group Arrest Study after Cardiac Hypothermia Rivers et al. Guérin et al. et al. Malmberg et al. García-Pagán Study Group MERIT et al. Herbrecht et al. Ramond Chen et al. et al. Johnson Williams Howie Liggins and 4 15 11 10 7 -HF 12 9 2 - 17 6 18 14 16 13 5 8 3 19 Cochrane bias risk* H(1,2) H(2,3) H(2,3) U3 U3 U3 U2 U2 U1 L L L L L L L L L Run in† N N N N N N N N N N N N N N N N Y Y centre‡ Multi- CMAJ N N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y | NOVEMBER 11,2019 Industry funded§ N N N N N N N Y Y Y Y Y Y Y Y Y Y Y LVEF <40% CHF andNYHA≥2, LVEF <35,NYHA3–4 symptoms within 24hof ST changes orLBBB CHF, NYHA4 (or riskofsuch) Trauma, hemorrhage Suspected meningitis sepsis syndrome sepsis, septic shockor Age ≥18yr, severe HE biopsy withDF>32or Alcoholic hepatitis on 24–36 wk Premature labour to transplantation and a contraindication oxygen consumption CHF, NYHA4,low peak ROSC within60min arrest from VForVT, Witnessed cardiac response to treatment ≥ 6PVC/h, LVEF <55%, MI last 2years, anticipated >3d Admitted to ICU for PMN >250 Age 18–80yr, ascitic for <36h Severe ARDSwithET invasive aspergillosis definite orlikely Immunocompromised, glucose >11.1mmol/L diabetes orrandom Acute MI,treated and VA drugs treatment withEGD variceal bleeding, Cirrhosis, acute Inclusion criteria | VOLUME 191 12.5–200 mg/d Metoprolol XL 25–50 mg/d Spironolactone 162 mg/d plusASA 75 mg/d Clopidogrel daily 2.5–20 mgtwice Enalapril 8 h 1 gIVbolus,over Tranexamic acid 6h 10 mgIVevery Dexamethasone 6 h resuscitation for Goal-directed for 28d 40 mg/d Prednisolone acetate 6mg betamethasone phosphate plus Betamethasone assist device Left ventricular 32–34°C for 1d Hypothermia flecainide Encainide or to 6.0mmol/L Target glucose 4.5 day3 day 1,1.0g/kg Albumin 1.5g/kg > 16h/d Prone ventilation 4–6 mg/kg/d Voriconazole 4 times/d subsequently 24 h,then IV insulinevery TIPS within72h | ISSUE 45 Intervention¶ Treatment duration ≥ 3min 12 wk 4.2 d 15 d 28 d 4 d 1 d 1 d 3 d 4 d 8 h 6 h ∞ ∞ ∞ ∞ – – Amphotericin B Normothermia “Conventional Target glucose ASA 162mg/d acetate 6mg ≤ 10mmol/L 1–1.5 mg/kg Endoscopic ventilation Control** treatment Cortisone therapy” Placebo Placebo Placebo Placebo Placebo Placebo Placebo Supine None None None ANALYSIS 1C E1245 -blinding PCI None None None Placebo Placebo Placebo Placebo Placebo 75 mg/d 7%–7.9% Control** Treatment Treatment Clopidogrel Clopidogrel targeting HbA targeting ASA 1300 mg/d Warfarin INR 2–3 Warfarin defibrillator, ICU = intensive care = intensive ICU -defibrillator, – – – ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 5 yr duration Treatment Treatment

1C up not cited or < 95%. cited -up not of follow -blinding, 3 = completeness lead ICD -lead Intervention¶ ISSUE 45 ISSUE | CABG Intensive Intensive treatment HbA targeting ICD INR 2–3 Warfarin Shock only, single Enalapril 2.5–10 mg twice daily Propranolol 40–80 mg 3 times/d Captopril 6.25–25 mg 3 times/d 90 mg Ticagrelor daily twice 5 mg Apixaban daily twice 10 mg/dRamipril < 6% Rosuvastatin 20 mg/d Tamoxifen

-MI 1C -CRP VOLUME 191 VOLUME | focusing on domains most relevant to this study (allocation concealment, double concealment, (allocation this study to relevant on domains most focusing -free 33 Inclusion criteria reactive protein, ICA = intracranial artery, ICD = implantable cardioverter ICD = implantable artery, ICA = intracranial protein, -reactive Diabetes, > 70% Diabetes, in ≥ 2 major stenosis vessels epicardial > 7.5%, high risk (Atrial fibrillation or fibrillation (Atrial + (age flutter) atrial ≥ 75 yr or previous or TIA or stroke embolism or systemic or CHF or diabetes hypertension) CAD, > 55 yr, Age PVD or stroke, and diabetes, (hypertension, total increased HDL, low cholesterol, smoking or cigarette microalbuminuria) type 2 40–79 yr, Age and HbA diabetes LDL < 3.4, HS > 2 mg/L cancer, Early breast 5 yr, tamoxifen disease Prior MI, LVEF < 30 < Prior MI, LVEF TIA or ≥ 40 yr, Age 50%–99% from stroke Rankin ICA stenosis, ≤ 3 score NYHA 2–3 ≥ 18 yr, Age < 35% LVEF CHF, < 35% LVEF CHF, MI acute 30–69 yr, Age LVEF 21–80 yr, Age < 40%, 3–16 d post within presenting ACS 24-h onset Y Y Y Y Y Y Y Y Y Y Y Y N funded§ Industry Industry NOVEMBER 11, 2019 NOVEMBER | -CRP = high-sensitivity C Y Y Y Y Y Y Y Y Y Y Y Y Y CMAJ Multi- centre‡ Y Y Y Y N N N N N N N N N in† Run = glycated hemoglobin, HS = glycated 1C L L L L L L L U2 U3 U2 U2 U2 H(2,3) bias risk* Cochrane Cochrane 25 32 31 22 29 23 20 30 28 26 27 21 24 Moss et al. Moss et Granger Granger al. et Bardy et al. et Bardy β-Blocker Attack Heart Trial et al. Pfeffer Wallentin al. et Farkouh al. et al. et Yusuf Gerstein al. et al. Davies et Chimowitz Chimowitz al. et al. et Yusuf Note: ACS = acute coronary syndrome, ARDS = adult respiratory distress syndrome, ASA = acetylsalicylic acid, CABG = coronary artery bypass graft, CAD = coronary artery disease, CHF = congestive congestive coronary CHF = artery CAD = disease, graft, artery acid, CABG = coronary bypass ASA = acetylsalicylic syndrome, distress ARDS = adult respiratory syndrome, coronary = acute ACS Note: tube, HDL = high-density lipoprotein, ET = endotracheal EGD = esophagogastroduodenoscopy, hepatitis), severity of alcoholic risk stratify to (an algorithm DF = discriminant function failure, heart HbA encephalopathy, HE = hepatic Study Table 1 (part 2 of 2): Summary of landmark randomized controlled trials in which treatments significantly reduced all-cause all-cause reduced significantly which treatments trials in controlled 2): Summary randomized landmark 2 of of 1 (part Table mortality density lipoprotein, LVEF = left ventricular ejection fraction, MI = myocardial infarction, N = infarction, MI = myocardial fraction, ventricular ejection = left LVEF block,-density lipoprotein, LDL = low LBBB = left bundle branch IV = intravenously, ratio, normalized unit, INR = international = premature PVC cells, intervention, coronary PMN = polymorphonuclear PCI = percutaneous disease, vascular PVD = peripheral Classification, Association Functional Heart NYHA = New York no, VF = ventricular = vasoactive, shunt, VA portosystemic intrahepatic = transjugular TIPS ischemic attack, TIA = transient circulation, of spontaneous ROSC = return contractions, ventricular or ongoing. ∞ = infinite or surgery, – = 1 procedure Y = yes, release, XL = extended tachycardia, VT = ventricular fibrillation, trials assessing risk of bias in randomized for tool Collaboration’s *Based on the Cochrane up). L = low risk, U = uncertain risk, H = high risk. 1 = no allocation concealment, 2 = no double risk, concealment, H = high risk. 1 = no allocation risk,-up). L = low U = uncertain of follow and completeness randomization. necessary for criteria by time followed observation had prerandomization †“Y” if study site. > 1 study at recruited were ‡“Y” if patients any company. financial support from cited §“Y” if study the intervention group. for ¶Treatment group. (or comparator) the control for **Treatment Ridker et al. et Ridker

ANALYSIS ment) foranotheryear. early breastcancerontamoxifen(afteraninitial5yearsoftreat­ between studies. standardized NNTsforanyoutcomethatonecarestocompare mula presentedabovecouldbemodifiedtocalculate1-year that expectedbychancealoneareimportant.Ifdesired,thefor­ ventions thatsignificantlychangethelikelihoodofdeathbeyond accurately byresearchers,makingitakeyRCToutcome.Inter­ mortality iscriticaltopatientsandalmostinvariablymeasured in these31studiesshouldallbeconsideredspecial.All-cause tions onpatientsurvival.However,theinterventionspresented into perspective and better gauge the importance of interven decreased all-causemortality.” that thecriticalreviewofRCTsshouldnotstopwith“significantly ventions’ influenceonpatientsurvival.Theseresultshighlight enced all-causemortality),largevariationsremainedintheinter­ nized landmarkRCTsofinterventionsthatsignificantlyinflu­ ourselves tothe“bestofbest”landmarkstudies(i.e.,recog­ E1246 ture labour with corticosteroids “being saved”annuallyaftertreatmentof8.3mothersinprema­ Chen et al. Group Trial Study CONSENSUS et al. Malmberg Johnson et al. Williams et al. Herbrecht van deBeek de Gansand Sort et al. Guérin et al. Pitt et al. Finfer et al. et al. García-Pagán Ramond et al. Rivers et al. Rose et al. Group Arrest Study after Cardiac Hypothermia Study Howie Liggins and results Table 2(part 1of 2):All-cause mortality outcomes inlandmarkrandomized controlled trialswithstatistically significant The 1-yeardeathNNTandTable2helpputRCToutcomes 7 15 10 12 9 2 - 17 6 14 16 13 5 8 3 11 4 Treatment, 22 961 10 060 3010 .* o n 127 137 157 144 130 237 126 822 306 61 32 63 32 32

Table 2showsthat,evenafterlimiting Control, Control, 22 891 10 067 3012 no.† 126 138 144 133 133 229 100 841 314 34 61 31 63 29 to continuing 556 women with CMAJ Treatment duration‡ ≥ 3min 12 wk 4.2 d 15 d 28 d 1 d 4 d 4 d 1 d 3 d 8 h 6 h ∞ ∞ – – | NOVEMBER 11,2019 Outcome time, yr§ 2.1667 1.3333 0.2466 0.0769 0.2308 0.1644 0.0767 0.9863 0.1538 0.0767 0.0192 0.2466 0.4932 0.5 1 2 Treatment** ­ 0.672 0.125 0.275 0.145 0.292 0.385 0.075 0.362 0.409 0.236 0.063 0.186 0.345 0.222 0.125 0.07 | report oflandmarkRCTs, 1991 and2003inTheClassicswereincludedIoannidis’seminal of keystudies;almostathirdthetrialspublishedbetween Hochman’s collectionofkeystudies tial onclinicalpractice. subjectively identifiedbyexpertstobeimportantandinfluen­ treatments hasitsdrawbacks.Weusedacollectionofstudies Any methodusedtosampleallpublishedstudiesevaluating example? What arethepotentiallimitationsofour comes. First,withtheexceptionofACEinhibitors vival andaddingthemtoourtable. that report a significant influence of treatment on all-cause ­ sur systematically identifyallothermethodologicallyrobustRCTs ment effectsonall-causesurvival.Furtherworkisrequiredto study mayhaveexcludedkeyRCTsreportingimportanttreat­ tolic dysfunction, results in our league table are based on a single between studiesincludedinThe Classics implantable cardioverter-defibrillators VOLUME 191 Death risk¶ Two issuesarisefromourresultsbeingbasedonRCTout­ Control†† 0.885 0.387 0.249 0.414 0.526 0.081 0.524 0.551 0.261 0.459 0.413 0.552 0.146 0.16 0.41 0.18 | ISSUE 45 ARR‡‡ –0.026 0.015 0.213 0.262 0.122 0.142 0.005 0.162 0.142 0.174 0.117 0.075 0.113 0.427 0.076 0.19 1 However,thereisconsiderableoverlap 35 value§§ < 0.001 < 0.001 0.0035 0.003 0.027 0.001 0.001 andmorethan40%ofRCTsin 0.01 0.03 0.02 0.02 0.03 0.03 0.02 0.02 0.04 p 36 wereinTheClassics. outcome 20,22 time¶¶ 184.2 –38.3 NNT- 67.6 13.4 13.2 4.7 3.8 5.3 8.2 7.1 6.2 5.7 8.6 8.8 2.3 7 1 forleftventricularsys­ andothercollections 0.16 (0.09to 0.64) NNT (95%CI)*** –9.5 (–64to –5.1) 1.3 (0.7to 7.9) 1.2 (0.6to 7.3) 0.4 (1.0to 2.7) 13 (7.1to 104) 1.2 (0.8to 2.3) 14 (7.4to 147) 5.2 (3.1to 16) 5.1 (2.1to 18) 1.9 (1.0to 23) 6.1 (3.5to 24) 3.5 (1.9to 20) 2.0 (1.1to 27) 1-year-death 10 (2.8to 14) 18 (12to 30) 12,23,25 1 Our and ANALYSIS E1247 28 (19 to 56) 28 (19 to 345 (99 to ∞) 345 (99 to 1-year-death 1-year-death 55 (45 to 185) 55 (45 to 428) 76 (42 to 81 (69 to 365) 81 (69 to 181) 85 (56 to 83 (46 to 480) 83 (46 to 30 (17 to 127) 30 (17 to 556 (272 to ∞) 556 (272 to First, both sta­ –334 (∞ to –54) –334 (∞ to 270 (159 to 881) 270 (159 to 156 (84 to 1084) 156 (84 to NNT (95% CI)*** –34 (–148 to –19) –34 (–148 to 38,39 . This dichoto­ 2. This Table 22 73.2 14.6 28.1 38.7 85.4 53.9 23.8 31.3 17.7 NNT- –18.5 –95.3 181.7 time¶¶ outcome outcome death number needed to harm). number needed to -death p 0.02 0.02 0.04 0.04 0.007 0.005 0.019 0.049 0.016 0.0036 < 0.001 < 0.005 < 0.001 value§§ –0.01 0.069 0.019 0.006 0.014 0.042 0.032 0.056 0.045 0.036 0.026 0.012 –0.054 ARR‡‡ ISSUE 45 ISSUE | Second, baseline death risks tend to exhibit secular Second, baseline death risks tend to exhibit secular 0.04 0.12 0.043 0.288 0.122 0.028 0.227 0.246 0.198 0.397 0.108 0.098 0.054 38 Control†† Death risk¶ Death VOLUME 191 VOLUME decreases over time as concomitant therapies and other aspects for example, 1-year mortality for of medical care improve; adults admitted to hospital in Ontario decreased by 20% between 1994 and 2009. This makes comparisons of NNT and Several potential limitations of the number needed to treat Several potential limitations of the number needed to treat (NNT) will also apply to the 1-year-death NNT. What are the potential limitations of using the 1-year-death NNT statistic? select treatments that were highlighted in treatments that were select mization of a continuous measure (i.e., the likelihood that differ­ ences seen exceed that arising by chance) will unfairly exclude some studies (such as those with a p value of 0.05) whose impor­ tance is not materially distinct from those presented here. tistics will tend to increase extensively when baseline risks tistics will tend to increase extensively when baseline risks decrease. | 0.22 0.05 0.097 0.104 0.022 0.214 0.204 0.088 0.142 0.352 0.073 0.072 0.043 Treatment** 5 1 1 3.5 1.9 7.6 7.6 3.5 1.85 3.45 This issue 1.6667 3.7917 2.0833 35 time, yr§ Outcome Outcome NOVEMBER 11, 2019 NOVEMBER | – – – ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 5 yr duration‡ Treatment Treatment CMAJ 953 490 280 847 no.† 4652 5123 8901 6440 2001 1921 1116 9291 1,284 Control,

37 947 742 289 829 n o .* 8901 6454 4645 5128 1916 1115 9333 1990 1,285 Treatment, Treatment, 19 25 32 31 22 29 23 20 -HF in groups. risk of death comparison for -sided p value 30 27 26 21 24 Other issues regarding our example should be kept in mind. Other issues regarding our example should be kept in mind. ¶¶Number needed to treat to prevent 1 death by outcome time. outcome by 1 death prevent to treat ¶¶Number needed to (the 1-year 1 year by 1 death cause to treat this is the number needed to negative, If 1 year. by 1 death prevent to treat ***Number needed to Note: ARR = absolute risk reduction, CI = confidence interval, NNT = number needed to treat, – = 1 procedure or surgery, ∞ = ongoing treatment throughout observation. observation. throughout treatment ∞ = ongoing or surgery, – = 1 procedure treat, NNT = number needed to interval, CI = confidence risk reduction, ARR = absolute Note: intervention group. assigned to *Number of people randomly group. control assigned to †Number of people randomly the intervention group. for treatment of active ‡Duration plus subsequent treatment active received it equals the time patients In closed studies, measured. or risk was count which death at randomization time is the time from §Outcome used. -up time was follow the average in study), patients between time varies outcome (i.e., risk. achieve observed In open studies to death time required observation risk. death maternal) (not ¶Neonatal assigned). randomly ÷ no. dead (no. in intervention group dead **Proportion assigned). randomly ÷ no. dead (no. group in control dead ††Proportion intervention group. risk for – death group control risk for as death calculated death, for risk reduction ‡‡Absolute §§Two Table 2 (part 2 of 2): All-cause mortality outcomes in landmark randomized controlled trials with statistically significant significant trials with statistically controlled in landmark randomized outcomes mortality 2): All-cause 2 of 2 (part Table results MERIT Group Study Bardy et al. et Bardy Study Yusuf et al. et Yusuf β-Blocker Attack Heart Trial Pfeffer et al. Pfeffer Farkouh al. et Gerstein al. et al. et Ridker al. Davies et Wallentin Wallentin al. et Yusuf et al. et Yusuf Chimowitz Chimowitz al. et Moss et al. Moss et analyses rather than could be addressed using data from meta-analyses rather than single studies. Second, the intervention effect measured in RCTs may generalize poorly. First, some of the health care technologies compared in our First, some of the health care technologies compared in our study are quite distinct, making their direct comparison relevant only in a numerical sense regarding the 1-year-death NNT. Sec­ was because this ond, we reported risk using a simple proportion reported in all studies. Risk-measurement methods that account such risk; measure accurately more will time observation for in calculable if proportions instead of be used can measurements the RCTs of all technologies that are to be compared. Finally, we used the p value for the intertreatment difference in death risk to

RCT of interventions. Single studies, even those that are large RCT of interventions. Single studies, even those that are large and internally valid, can return exaggerated results that can be or are replicated trials even absent) when sometimes (or smaller when the intervention is applied in clinical practice. ANALYSIS E1248 1-yeardeath NNTbetweenyearsproblematic. meas­ures. deaths usinghealthutilitiesand-relatedqualityoflife ure). Furtherresearchmightexplorethepossibilityofweighting someone dying with progressive, incapacitating respiratory fail­ worse thanothers(e.g.,contrastapersondyingintheirsleepto This isarguablynottruesincesomedeathsaresymptomatically technology comparisons,weassumedthatalldeathsareequal. in usingall-causemortalitytopermitcrossstudyand treatment effects. mate deathriskinstudieswithlongobservationtimesandlarge This madeinterstudycomparisonspossiblebutcouldunderesti­ recalculation, deathrisksweresummarizedusingproportions. not usesurvivalanalysisorprovidethenecessarydatatopermit from randomizationtooutcomemeasurementincreases. reciprocal oftheabsoluteriskreduction)willdecreaseastime larger the longer patients are observed. Therefore, the NNT (the estimates —anddifferencesbetweentreatmentgroupsget weighting 1 health thandeathinanolderadult.Furtherworkcouldexplore mature labour is more consequential with regard to population for life years lost owing to the death. Neonatal death during pre­ the 1 in eachpatientgrouptotrulyappreciatetreatmenteffects.Fifth, Therefore, one should always examine the absolute death risks lute deathriskwithpotentiallyimportantlossofinformation. Fourth, the1-yeardeathNNTsimplifiescomparisonofabso­ 2 yr),therebyreducingtheinfluenceofthispotentialbias. analyzed (n estimate. Theoutcometimesofalmosthalfthestudieswe nificantly overtime,the1-yeardeathNNTmayreturnabiased study’s reported outcome time. If treatment effect changes sig death riskisthesame1yearafterrandomizationasat dardization assumesthattherelativeinfluenceoftreatmenton time framesto1yearafterrandomization.However,thisstan­ 1-yeardeath NNTavoidsthislimitationbystandardizingall one shouldalways examineitsindividualdatacomponents. understand the1-yeardeathNNT(aswithanysummary statistic), value. Thispossibilitysupports thepointmadeabove:totruly occurred, 1-yeardeathNNTwouldreturnaninappropriately large tion timethatnotablyexceeded theperiodduringwhichdeaths will beincreasinglyflatovertime. Ifsuchastudychoseanobserva­ close to the start ofobservation;thecorrespondingsurvival curves which deathsinbothtreatmentgroupsareprimarilyclustered with theoutcometimechoseninRCT.Considerastudy in should alsobeconsidered.Finally,the1-yeardeathNNTwillvary nologies. Otherfactors,suchascostanddiseaseprevalence, is butonestatisticthatcouldbeusedtocomparehealthcaretech ­ out influencingthelikelihoodofdeath.Fourth,1-yeardeathNNT interventions mightprovideimportantbenefitstopatientswith ­ ysis techniques. most accuratewhencensoringisconsideredusingsurvivalanal­ having variableobservationtimeforpatients,deathriskswere values could varywidely in particularsubgroups. Finally, in RCTs measures) representtheaveragevalueofpatientsinthosetrials; year death NNT. First, Several limitationsareparticulartothe1-yeardeathNNT.First, - year- 42 death NNTs reported in Table 2 (just as with all NNT death NNT also does not account Second,the1-yeardeathNNTalsodoesnotaccount =15)wereverycloseto1year(between6moand - year 41 - Since most RCTs in our worked example did death NNT by expected life years lost. Third, CMAJ | NOVEMBER 11,2019 40 Third,deathrisk 38 The ­ |

10. 14. 13. 12.

References distinct treatmentoptions. or fundingandhelpindividualphysiciansdiscriminatebetween health careorganizationsprioritizetechnologiesforevaluation potential impactsofhealthcareinterventions.Itcouldalsohelp statistic to help consumers of medical evidence contextualize Despite its limitations, the 1-yeardeath NNT represents a useful Conclusion 15. 11. 19. 17. 16. 18. VOLUME 191 9. 1 5 8 3 6 2 4. 7...... Rivers patients withseverealcoholichepatitis.NEnglJMed1992;326:507-12. in severeacuterespiratorydistresssyndrome. Sort shock. Group. Ramond de Gans for primarytherapyofinvasiveaspergillosis.NEnglJMed2002;347:408-15. and theGlobalAspergillusStudyGroup. Group oftheEuropeanOrganisationforResearchandTreatmentCancer García-Pagán 549-56. to improvetheneurologicoutcomeaftercardiacarrest. Hypothermia after Cardiac Arrest Study Group. gitis. Meningitis StudyInvestigators. Guérin atrics for preventionoftherespiratorydistresssyndromeinprematureinfants. Rose 360:​ NICE-SUGAR StudyInvestigators; Survival Study(CONSENSUS).NEnglJMed1987;316:1429-35. gestive heartfailure.ResultsoftheCooperativeNorthScandinavianEnalapril CONSENSUS TrialStudyGroup. Herbrecht tonitis. ment andmortalityinpatientswithcirrhosisspontaneousbacterialperi­ Liggins of thelandmarktrials.1sted.Boston:2MinuteMedicineInc.;2015 Succi Malmberg ure. Group. Assistance fortheTreatmentofCongestiveHeartFailure(REMATCH)Study versus conventional glucose control in critically ill patients. with cirrhosisandvaricealbleeding.NEnglJMed2010;​ Portosystemic Shunt)CooperativeStudyGroup. Williams-Johnson Intervention TrialinCongestive HeartFailure(MERIT-HF). Effect ofmetoprololCR/XLin chronic heart failure: Metoprolol CR/XL Randomised Trial. receiving encainide,flecainide,orplacebo. TheCardiacArrhythmiaSuppression and mortalityinpatientswithsevereheart failure. Pitt controlled trial. rin in45,852patientswithacutemyocardialinfarction:randomisedplacebo- Myocardial InfarctionTrial)collaborativegroup. Chen Cardiol myocardial infarction(DIGAMIstudy):effectsonmortalityat1year. sion followedbysubcutaneousinsulintreatmentindiabeticpatientswithacute Echt controlled trial. patients withsignificanthaemorrhage(CRASH-2):arandomised,placebo acid ondeath,vascularocclusiveevents,andbloodtransfusionintrauma 1283- B N EnglJMed P N EnglJMed DS, EA, N EnglJMed MD, , 1972; ZM, , E N EnglJMed Zannad C Navasa N EnglJMed directed therapy in the treatment of severe sepsis and septic Early goal-directedtherapyinthetreatmentofseveresepsisandseptic Long-term useofaleftventricularassistdeviceforendstageheartfail­ 1995; GC, , ,

Nguyen J Gelijns MJ Reignier Liebson Carr 97. , Jiang K 50: Howie R van deBeek , , 26: , Rydén Poynard JC, LH, | 515- Denning ISSUE 45 M F 57- Lancet West IndianMedJ2010;59:612-24. AC, , , LX, 2001; B Caca Cheung 2002; Remme Arroyo 1991; RN. J 25. PR, , 65. JA, , 2001; Havstad L 1999; Moskowitz Richard Chen ,

A controlled trial of antepartum glucocorticoid treatment A controlledtrialofantepartumglucocorticoidtreatment 345: Efendic McDonald Mitchell T 347: 2005; K, 324: DW, , 345: V D 341: A Rueff Bureau WJ, , 1435- ; YP, . 1549- et al. 781- 2 Minute Medicine’s the classics in medicine: summaries 2 MinuteMedicine’stheclassicsinmedicine:summaries European DexamethasoneinAdulthoodBacterial 366: JC, S 1368- Patterson 403- , S et al. et al.;

et al.; AJ, Dexamethasone inadultswithbacterialmenin­ B , LB, 43. 8 et al.; Effects ofenalaprilonmortalityinseverecon­ 56. 1607- Effect of intravenous albuminonrenalimpair­ et al. , . AH, C, 9 77. Finfer et al. et al.; . et al. et al.; The effect of spironolactone on morbidity The effectofspironolactoneonmorbidity Directed Therapy Collaborative Early Goal-DirectedTherapy Collaborative COMMIT (ClOpidogrelandMetoprololin Strachan 21. PROSEVA StudyGroup. Randomized trialofinsulin-glucoseinfu­ TF, A randomized trial of prednisolone in S, Randomized Evaluation of Mechanical Randomized EvaluationofMechanical Mortality andmorbidityinpatients Early TIPS(Trans­ Voriconazole versusamphotericinB Chittock et al.; N EnglJMed GG, Addition ofclopidogrel to aspi­ Mild therapeutic hypothermia N EnglJMed Early useofTIPSinpatients Invasive FungalInfections et al. DR, 362:2370-9. Lancet N Engl J Med Su Effects oftranexamic 2013; jugular Intra­ SY, N EnglJMed Prone positioning Prone positioning 1999; 1999; 368:​ et al. 353: 341:​ 2159- 2002;346:​ J Am Coll J AmColl Intensive 2001- 709- hepatic 68. 2009;​ Pedi- 17. 7 . ANALYSIS

54. : 1 E1249 16. 2003; Rosuvastatin 805- Control Clin Trials Biostat Epidemiol 381:​ 2013; 37. Lancet 225- ; Cochrane Bias Methods Group 352: . Study Group JUPITER Number needed to treat is incorrect Health Qual Life Outcomes et al.; The Cochrane Collaboration tool for tool for The Cochrane Collaboration 2005; The Health Utilities Index (HUI): concepts, et al.; et al. PC, B, Sudden Cardiac Death in Heart Failure TrialSudden Cardiac Death in Heart Adjuvant Tamoxifen: Longer Against ShorterAdjuvant Tamoxifen: FAH, et al. , year survival of people admitted to hospital in in to hospital admitted of people survival year ­ of continuing adjuvant tamoxi Long-term effects D et al.; et al.; 207. , N Engl J Med Goetzsche J Amiodarone or an implantable cardioverter-defibrillator Amiodarone or an implantable DB, Fonseca Feeny , 2195- , E DG, Smiechowski W Mark P, 359: Trends in 1- in Trends Godwin . , KL, ISSUE 45 ISSUE C Contradicted and initially stronger effects in highly cited clinical Contradicted and initially stronger The Kaplan–Meier method for estimating and comparing pro­ The Kaplan–Meier method for estimating and comparing 50 studies every doctor should know: the key studies that form the form that studies key the know: should doctor every studies 50 H |

The “number needed to treat” turns 20 — and continues to be to be The “number needed to treat” turns 20 — and continues Altman . . 2008; Furlong 23. Brett Hryciw and Carl van Walraven contributed to the to contributed Walraven van Carl and Hryciw Brett Danielson Lee SX. , 33. ME Generalizing the results of randomized clinical trials. Generalizing the results of randomized FA. Pan JPA. J

Brassard , JPT, 15- 23- C Xie D, PM, : GH, KR. , 2 15: J measurement properties and applications. Ontario, 1994–2009. CMAJ 2013;185:E755-62. Zee . Revised ed. Oxford (UK): Oxford Universitymedicine. Revised ed. Oxford (UK): Oxford based evidence of foundation Press; 2013. Bailey 2012;65:42-6. without proper time-related considerations. J Clin Epidemiol Walraven van portions in a randomized controlled trial with dropouts. 2018;​ Horsman Higgins Ioannidis -28. research. JAMA 2005;294:218 Hochman 1994; McAlister used and misused. CMAJ 2008;179:549-53. Suissa (SCD-HeFT) Investigators. for congestive heart failure. Cochrane Statistical Methods Group. Cochrane Statistical Methods trials. BMJ 2011;343:d5928. assessing risk of bias in randomised Bardy Ridker (ATLAS) Collaborative Group. (ATLAS) Collaborative receptor- after diagnosis of oestrogen versus stopping at 5 years fen to 10 years trial. cancer: ATLAS, a randomised positive breast N Engl J Med to prevent vascular events in men and women with elevated C-reactive protein. events in men and women to prevent vascular Davies VOLUME 191 VOLUME 41. 37. 40. 42. 33. 35. 36. 38. 39. Contributors: Carl van Walraven and conception and data collection. Brett Hryciw, of the authors contrib­ Meltem Tuna contributed to the data analysis. All and Finlay McAlisteruted to the data interpretation. Carl van Walraven All of the authorsdrafted the article, which all of the authors reviewed. and agreed to be gave final approval of the version to be published accountable for all aspects of the work. Correspondence to: Carl van Walraven, [email protected] 34. 31. 32. |

P, HC, 83. N Engl J 877- Sleight 346: S, Gerstein 2002; Yusuf NOVEMBER 11, 2019 NOVEMBER | Effect of enalapril on survival ARISTOTLE Committees and ARISTOTLE Committees and N Engl J Med Aspirin Symptomatic Warfarin-Aspirin Symptomatic . FREEDOM Trial Investigators N Engl J [published erratum in N Engl J CMAJ et al. et al.; et al.; CE, et al.; Comparison of warfarin and aspirin for aspirin and warfarin of Comparison H, Effect of captopril on mortality and mor­ and mortality on captopril of Effect JJV, Ticagrelor versus clopidogrel in patients Ticagrelor versus clopidogrel LA, Davis , - an implantable cardioverter Amiodarone or et al. et 302. B Multicenter Automatic Defibrillator Implantation Defibrillator Automatic Multicenter et al. LA, A, 293- Pitt Effects of intensive glucose lowering in type 2 dia­ Sleeper et al. , McMurray et al.; , S 77. M 325: Moyé DB, Howlett-Smith Budaj JH, et al. , WJ, E 669- Yusuf Prophylactic implantation of a defibrillator in patients with of a defibrillator Prophylactic implantation MJ, 1991; RC, RP, Hall Mark 327: , W Apixaban versus warfarin in patients with atrial fibrillation. atrial with patients in warfarin versus Apixaban KL, Effects of an angiotensin-converting enzyme inhibitor, ramipril, Lynn Domanski Alexander 1992; Becker MI, Braunwald Lee Byington L, ME, et al. CB, Zareba N Engl J Med MA, J, GH, ME, AJ, A randomized trial of propranolol in patients with acute myocardial infarction: A randomized trial of propranolol ;247:1707-14. I. mortality results. JAMA 1982 Pfeffer in patients with reduced left ventricular ejection fractions and congestive heartin patients with reduced left failure. -37. N Engl J Med 2005;352:225 congestive heart failure. defibrillator for SOLVD Investigators; Chimowitz Moss Investigators. Trial Disease Intracranial -16. N Engl J Med 2005;352:1305 intracranial arterial stenosis. symptomatic Bardy . Trial II Investigators fraction. and reduced ejection myocardial infarction bidity in patients with left ventricular dysfunction after myocardial infarction.bidity in patients with left enlargement trial. The SAVE Investigators. Results of the survival and ventricular Pogue N Engl J Med with diabetes. Strategies for multivessel revascularization in patients Med 2012;367:2375-84. Granger . Investigators Wallentin Miller betes. N Engl J Med 2008;358:2545-59. on cardiovascular events in high-risk patients . Med 2000;342:748, 1376]. N Engl J Med 2000;342:145-53 Group; Action to Control Cardiovascular Risk in Diabetes Study N Engl J Med 2009;361:1045-57. with acute coronary syndromes. Farkouh N Engl J Med 2011;365:981-92. ; Heart Outcomes Prevention Evaluation Study Investigator Competing interests: None declared. This article has been peer reviewed. The Walraven), van (Hryciw, Medicine of Department Affiliations: Internal Medicine Ottawa Hospital, Ottawa, Ont.; Division of General ICES uOttawa (Tuna,(McAlister), University of Alberta, Edmonton, Alta.; (Tuna, van Walraven),van Walraven); Ottawa Hospital Research Institute Ottawa, Ont. 24. 25. 23. 21.

20. 22. 28. 26. 30. 27. 29.