Yo Your ur Fi Findings dings Leo Buckley, ley, PharmD mD - PowerPoint PPT Presentation

Research search In Inside ide and d Ou Out: t: How w to to Ask th the Rig ight ht Qu Question, estion, Ge Get t th the Best t Answer swer and d Publish blish Yo Your ur Fi Findings dings Leo Buckley, ley, PharmD mD Brigham and Women’s Hospital Boston, ton, MA

Di Disc sclosur losure: e: This presentation is based upon available evidence, published expert opinion and my own personal anecdotes

Ou Outcome tcomes s of f Reside sidency ncy Research search Projects jects Median Time to Publication of Resident Publication Rates of Research Residency Abstracts 0 5 10 15 20 25 Months Published Unpublished IRB Data Collection Analysis Publication McKelvey, AJHSP 2010 Olsen, AJHSP 2012

Lim imitin iting g Fa Factors tors Associated ociated wit ith h Resident sident Resea search rch • Selection (design) of a project unsuitable for publication • Lack of knowledge of the publication process • Manuscript rejection • Time constraints • Limited mentorship for the trainee • Lack of trainee interest in a project • Lack of perceived benefit associated with publishing • Financial constraints for statistical support or publication fees • Trainee departs without completing manuscript Deal, Pharmacotherapy 2016

Se Sele lection ction (d (desig esign) n) of of a pro a project ject un unsui suitable table fo for pu r publ blicatio ication

Matching tching th the Siz ize e of Yo f Your ur Qu Question stion to to th the Siz ize e of f Your ur Answer wer • Does Drug A reduce mortality? • Focus on a specific knowledge gap • Helpful framework: write down the question and your hypothesized answer o Does the use of an IV furosemide dosing algorithm lead to consistent diuretic response in patients with heart failure and congestion? o We hypothesize that diuretic response, as measured by 3-hour urine output, is similar across patients with varying characteristics and oral diuretic usage.

Envisioning isioning what at your r stu tudy dy wil ill l produce oduce • You should be able to write the abstract, tables and figures before starting the study.

Wh What at is statist atistical ical po powe wer? • Definition: power = 1 – β where β is the probability of concluding that there is no significant effect, when in fact, there is a real effect i.e., your drug worked, but your statistics failed so, power is the probability of finding the effect that you think exists

Wh Why y do I ne I need ed to to est stimate imate sa sample mple si size ze? ~17.5% absolute difference!!! Buckley Curr Emerg Hosp Med Rep 2016

Wh Why do I n I need ed to to esti timate mate sampl mple e siz ize? e? • Assuming a 30% rate in the bridge group and a 12.5% rate in the no bridge group (which is a huge difference), enrollment of 38 participants in the no bridge group and 17 in the bridge group provides about 32% power to detect a significant difference. • So, if there really is a 17.5% difference between the bridge and no bridge groups ( which is very unlikely anyway ), then my study as designed would detect the 17.5% difference about 1 in every 3 times on average.

Ho How w do do I es I estimate imate sam ampl ple e size? ze? 1. Recall that sample size estimations are estimates 2. Guess the average in the treatment group 3. Guess the average in the control group 4. Guess the variability of the between-group difference in the treatment and control group 5. Plug your estimates into at least TWO sample size estimators (powerandsamplesize.com) 6. Repeat using different averages and variabilities

An An ex exam ample ple po powe wer es estimati imation. on. in a Randomized Clinical Trial Between-Group GLS Change (%) SD -2.2 -2.5 -2.7 2.7 85% 92% 95% 3.0 78% 87% 92% 3.5 65% 76% 82% CRP=C-reactive protein; SD

What if I can’t enroll the optimal number of f subjects? jects? • Change your endpoint: continuous endpoints are more sensitive to change than categorical endpoints o Change in hemoglobin • Combine categorical endpoints (30%  65% power): o Major bleeding = ~20% of patients o Major or clinically relevant nonmajor bleeding = ~35% of patients o Net clinical benefit (major or clinically relevant nonmajor bleeding or any stroke or systemic embolism) = ~40% of patients • Use a case-control study design

Case-Contro Ca Control l St Stud udies es Controls Cases Exposure(+) Exposure(-) Exposure(+) Exposure(-) N=50 N=100 N=200 N=50 Odds of exposure for cases = 200 / 50 = 4:1 Odds of exposure for controls = 50 / 100 = 1:2 Odds ratio for cases vs. controls = (4/1) / (1/2) = 8

Co Coho hort t St Stud udies ies N=125 Cases Probability=125/200 Exposure(+) Controls N=200 Relative risk=125/200 All Patients 20/100 of Interest N=20 Cases Probability=20/100 Exposure(-) Controls N=100

How w to to decide ide between tween a case-cont control rol stu tudy dy and d a cohort hort stud tudy? y? Case-Control Cohort • Good for rare outcomes • Bad for rare outcomes • Good for outcomes that occur • Your exposed and unexposed years after exposure groups are derived from the same overall sample • Very feasible • Can estimate disease incidence • Who is in your control group? • You need a quick answer (i.e., outbreaks) • Unsure of the disease incidence

Wh Which h statist atistical ical tes est t do do I us I use? e? Alternatively, could use regression models to adjust for confounders Slide courtesy of William Baker PharmD, UConn

Wh Wher ere e do does es the he da data a com ome e from om?

A d ju d ic a tio n o f S e lf-R e p o rte d F e m a le A d ju d ic a tio n o f S e lf-R e p o rte d W h ite P a rtic ip a n ts in H e a lth L N K P a rtic ip a n ts in H e a lth L N K 4 0 8 4 0 0 3 4 4 4 0 0 79 3 0 0 21 2 0 N u m b e r N u m b e r 5 0 1 0 4 5 12 1 1 2 0 0 F e m a le M a le O th e r M is s in g W h ite As ia n B la c k H is p a n ic O th e r M is s in g S e x in H e a lth L N K R a c e /E th n ic ity in H e a lth L N K Ahmad A, et al. Circulation 2017;136:1207-16.

A v a ila b ility o f H y p e rte n s io n IC D -9 C o d e s , A n ti-H y p e rte n s iv e M e d ic a tio n U s e a n d B lo o d P re s s u re M e a s u re m e n ts in H e a lth L N K 1 5 0 Good! Bad! Huge problem! 1 0 0 N u m b e r 5 0 0 IC D -9 , M e d s M is s in g M is s in g M is s in g IC D -9 o n ly M e d s o n ly B P o n ly a n d B P IC D -9 M e d s B P D a ta A v a ila b ility Ahmad A, et al. Circulation 2017;136:1207-16.

Hogan 1997 JAMIA Chan 2010 Med Care Res Rev

Potential tential Probl oblems ems wit ith h Mic icrosoft rosoft Excel cel • No tracking o Excel does not care if you delete a value by accident o Excel does not care if you sort the data incorrectly o Excel does not care if you copy and paste incorrectly o Excel does not care if you add a column for new data and delete another o Excel does not care if you enter “1300” for systolic blood pressure o Excel does not care about missing data

Using ing Mic icrosoft rosoft Excel el fo for Data ta Coll llection ection • Create a new version each time you open your Excel spreadsheet and track what you did each time and who did it

Using ing Mic icrosoft rosoft Excel el fo for Data ta Coll llection ection • Always double check Excel after you ask it to do something o Sorting rows – did you select all columns or just some? o Performing calculations – did you select the right area? Did you use the right command?

Using ing Mic icrosoft rosoft Excel el fo for Data ta Coll llection ection • Use specific terms for missing data vs. data that you haven’t collected yet o “NA” = I looked and it’s not there

Cl Clea ean n you our da data • Check the 5 or 10 highest and lowest values for outliers, check for missing data, spot check cases to make sure you didn’t sort incorrectly

Da Data a Co Collection ection So Software ware

La Lack ck of of kn knowledge owledge of of th the e publ pu blicati ication on pr proces ocess Ma Manuscript nuscript re rejection ection

Au Autho horship rship • Why is it important? o Getting credit – people want to be authors o Taking responsibility – people have to be authors

Au Autho horship rship • First author: person who does the bulk of the work • Second author: person who does a lot of the work • Middle authors: people who contributed deeply in specific areas or broadly to the overall study • Last author: senior author, person who provided oversight and direction of the study, group leader or principal investigator • Corresponding author: someone who can speak on behalf of the authors