1 00:00:04,738 --> 00:00:09,008 GOOD EVEN 2 00:00:09,075 --> 00:00:16,316 EVERYONE, MY NAME IS TEE YAN CHA 3 00:00:16,383 --> 00:00:24,257 WU, MY EXPERTISE IS THE 4 00:00:24,324 --> 00:00:25,825 STATISTICAL MONITORING. IN THIS 5 00:00:25,892 --> 00:00:28,995 LECTURE, I AM GOING TO PRESENT 6 00:00:29,062 --> 00:00:30,697 STATISTICAL POWER ANALYSIS. 7 00:00:30,764 --> 00:00:32,532 FIRST, I WILL INTRODUCE 8 00:00:32,599 --> 00:00:35,935 DEFINITION OF STATISTICAL POWER 9 00:00:36,002 --> 00:00:39,939 THREE COMMONLY USED STUDY 10 00:00:40,006 --> 00:00:42,642 DESIGNS AND THE CALCULATION. 11 00:00:42,709 --> 00:00:43,176 TODAY WE WILL DISCUSS THE 12 00:00:43,243 --> 00:00:43,743 FACTORS AFFECTING THE POWER 13 00:00:43,810 --> 00:00:48,515 ANALYSIS AND LATER I WILL 14 00:00:48,581 --> 00:00:52,385 PRESENT TWO EXAMPLES OF POWER 15 00:00:52,452 --> 00:00:56,756 ANALYSIS. WITHOUT THIS, IN THIS 16 00:00:56,823 --> 00:00:59,359 LECTURE, THE OUTCOME VARIABLES 17 00:00:59,426 --> 00:01:03,096 ARE STILL TO BE CONTINUES AND 18 00:01:03,163 --> 00:01:04,831 NORMALLY DISTRIBUTED AND THERE'S 19 00:01:04,898 --> 00:01:07,267 A TREATMENT AND CONTROL GROUPS 20 00:01:07,333 --> 00:01:07,834 HAVE -- THE SAME OR SIMILAR 21 00:01:07,901 --> 00:01:12,939 VARIABLES. FIRST I WOULD LIKE 22 00:01:13,006 --> 00:01:15,375 TO INTRODUCE TWO TYPE OF 23 00:01:15,442 --> 00:01:20,313 STUDIES. THAT WE DON'T WANT. 24 00:01:20,380 --> 00:01:24,184 WHY IS UNDER POWER STUDY WHICH 25 00:01:24,250 --> 00:01:26,820 HAS A SAMPLE SIZE TOO SMALL. 26 00:01:26,886 --> 00:01:30,190 BECAUSE OF THE SMALL SAMPLE SIZE 27 00:01:30,256 --> 00:01:35,361 IT USES THE WILD DIFFUSIONS FOR 28 00:01:35,428 --> 00:01:38,798 THE LIMITS FOR EXAMPLE, IF THE 29 00:01:38,865 --> 00:01:40,200 SLIDES OR SAMPLE MEANS MAY HAVE 30 00:01:40,266 --> 00:01:43,903 A LARGE BARRIERS OR ERROR 31 00:01:43,970 --> 00:01:46,206 THEREFORE IT DOES NOT HAVE TO 32 00:01:46,272 --> 00:01:47,774 THE NECESSARY POWER TO DETECT 33 00:01:47,841 --> 00:01:50,076 THE TRUE DIFFERENCE, LEADING TO 34 00:01:50,143 --> 00:01:55,348 A FALSE NEGATIVE CONCLUSION. IT 35 00:01:55,415 --> 00:01:57,717 EXPOSE Z PARTICIPANTS TO RISK 36 00:01:57,784 --> 00:01:58,318 WITHOUT PROVIDING MEANINGFUL 37 00:01:58,384 --> 00:02:06,826 KNOWLEDGE. AND IT ALSO LEAD TO 38 00:02:06,893 --> 00:02:10,230 BIAS CONCLUSION AND LOW 39 00:02:10,296 --> 00:02:12,098 REPRODUCIBILITY OF RESULTS. THE 40 00:02:12,165 --> 00:02:16,135 OTHER ONE IS OVERPOWERED STUDY 41 00:02:16,202 --> 00:02:17,704 WHICH HAS SAMPLE SIZE TOO LARGE 42 00:02:17,770 --> 00:02:20,607 IT MAY RESULT IN STATISTICALLY 43 00:02:20,673 --> 00:02:21,274 SIGNIFICANT DIFFERENCES THAT ARE 44 00:02:21,341 --> 00:02:25,745 NOT CLINICALLY IMPORTANT. FOR 45 00:02:25,812 --> 00:02:27,380 EXAMPLE WITH A LARGE SAMPLE SIZE 46 00:02:27,447 --> 00:02:31,384 THE STUDY MAY FIND A CHANGE IN 47 00:02:31,451 --> 00:02:34,220 THE I 1C OF 0.1% WHICH IS NOT 48 00:02:34,287 --> 00:02:43,162 CLINICALLY IMPORTANT. IT EXPOSE 49 00:02:43,229 --> 00:02:43,830 Z AN UNNECESSARY NUMBER OF 50 00:02:43,897 --> 00:02:45,331 PARTICIPANT TO RISK BOTH 51 00:02:45,398 --> 00:02:46,266 UNDERPOWERED AND OVERPOWERED 52 00:02:46,332 --> 00:02:53,473 STUDIES WASTE RESOURCES. SO DO 53 00:02:53,540 --> 00:02:56,376 WE NEED A POWER ANALYSIS? THE 54 00:02:56,442 --> 00:02:59,112 REASON IS TO AVOID THE STUDY 55 00:02:59,178 --> 00:03:01,848 FROM BEING UNDERPOWERED OR 56 00:03:01,915 --> 00:03:04,217 OVERPOWERED THEREFORE TO ENSURE 57 00:03:04,284 --> 00:03:05,852 THE STUDY HAS SUFFICIENT 58 00:03:05,919 --> 00:03:09,355 STATISTICAL POWER TO DETECT 59 00:03:09,422 --> 00:03:12,058 EFFECTS THAT ARE SCIENTIFICALLY 60 00:03:12,125 --> 00:03:15,094 IMPORTANT AND CLINICALLY UNIFORM 61 00:03:15,161 --> 00:03:17,130 AND TO ALLOCATE SUFFICIENT BUT 62 00:03:17,196 --> 00:03:17,797 NOT EXCESSIVE RESOURCES TO DATA 63 00:03:17,864 --> 00:03:22,201 COLLECTION. POWER ANALYSIS HAS 64 00:03:22,268 --> 00:03:25,204 TWO ROLES. ONE IS TO DETERMINE 65 00:03:25,271 --> 00:03:26,973 THE NECESSARY NUMBER OF SUBJECTS 66 00:03:27,040 --> 00:03:31,411 NEEDED TO DETECT A GIVEN EFFECT 67 00:03:31,477 --> 00:03:34,380 SIZE OR SOME SITE ESTIMATION. 68 00:03:34,447 --> 00:03:36,716 THE OTHER ROLE IS TO DETERMINE 69 00:03:36,783 --> 00:03:38,117 THE POWER GIVEN THE EFFECT SIZE 70 00:03:38,184 --> 00:03:39,352 AND THE NUMBER OF SUBJECTS 71 00:03:39,419 --> 00:03:47,193 AVAILABLE OF POWER ESTIMATION. 72 00:03:47,260 --> 00:03:52,231 SO WE -- HAVE ANALYSIS IS 73 00:03:52,298 --> 00:03:57,804 PERFORMED. IN MOST TIMES, POWER 74 00:03:57,870 --> 00:03:58,871 IS ANALYSIS IS PERFORMED BEFORE 75 00:03:58,938 --> 00:04:03,476 THE DATA IS COLLECTED OR DURING 76 00:04:03,543 --> 00:04:05,678 PROTOCOL DEVELOPMENT. THE 77 00:04:05,745 --> 00:04:07,080 PURPOSE IS TO ESTIMATE A SAMPLE 78 00:04:07,146 --> 00:04:08,748 SIZE OR TO DETERMINE THE POWER 79 00:04:08,815 --> 00:04:11,351 FOR GIVING SAMPLE SIZE. A POWER 80 00:04:11,417 --> 00:04:13,586 ANALYSIS IS REQUIRED FOR THE 81 00:04:13,653 --> 00:04:19,359 PROTOCOLS WITH STATISTICAL 82 00:04:19,425 --> 00:04:21,995 HYPOTHESES. POWER ANALYSIS IS 83 00:04:22,061 --> 00:04:24,764 ALSO PERFORMED IN INTERIM 84 00:04:24,831 --> 00:04:26,432 ANALYSIS. THE PURPOSE IS TO 85 00:04:26,499 --> 00:04:28,434 REESTIMATE THE SAMPLE SIZE BASED 86 00:04:28,501 --> 00:04:32,171 ON REESTIMATED COMPONENTS SUCH 87 00:04:32,238 --> 00:04:34,207 AS STANDARDIZATION AND EFFECT 88 00:04:34,273 --> 00:04:36,442 SIZE ALL TO CALCULATE THE 89 00:04:36,509 --> 00:04:39,345 CONDITIONAL POWER USED FOR 90 00:04:39,412 --> 00:04:45,418 FUTILITY STOPPING. SOMETIMES 91 00:04:45,485 --> 00:04:46,819 POWER ANALYSIS IS EFFECTED AFTER 92 00:04:46,886 --> 00:04:48,521 THE DATA IS COLLECTED. THE 93 00:04:48,588 --> 00:04:52,225 PURPOSE IS TO CALCULATE THE 94 00:04:52,291 --> 00:04:54,927 POWER OR TO USE THE SAMPLE SIZE 95 00:04:54,994 --> 00:04:58,231 FOR THE FUTURE BASED ON THE 96 00:04:58,297 --> 00:05:03,369 COMPONENTS. FOR EXAMPLE, A 97 00:05:03,436 --> 00:05:07,540 STUDY OF THE DIFFERENCE BETWEEN 98 00:05:07,607 --> 00:05:10,209 THE TWO GROUP IS CLINICALLY 99 00:05:10,276 --> 00:05:13,579 IMPORTANT. BUT NOT AS 100 00:05:13,646 --> 00:05:14,280 SPECIFICALLY SIGNIFICANT YOU MAY 101 00:05:14,347 --> 00:05:18,017 CHECK THE POWER. YOU MAY CHECK 102 00:05:18,084 --> 00:05:19,352 THE POWER ANALYSIS COULD CHECK 103 00:05:19,419 --> 00:05:22,221 IN THE STUDY HAS ENOUGH POWER OR 104 00:05:22,288 --> 00:05:23,690 TO ESTIMATE A SAMPLE SIZE FOR 105 00:05:23,756 --> 00:05:34,300 FUTURE STUDY. WE INTRODUCE TWO 106 00:05:38,304 --> 00:05:40,039 TYPES OF HYPOTHESIS AND ERRORS, 107 00:05:40,106 --> 00:05:43,576 ONE IS CALLED NULL HYPOTHESIS 108 00:05:43,643 --> 00:05:47,346 WHICH REPRESENTS WHAT THE 109 00:05:47,413 --> 00:05:48,448 INVESTIGATOR AIMS TO DISPROVE. 110 00:05:48,514 --> 00:05:51,350 THE OTHER ONE IS CALLED 111 00:05:51,417 --> 00:05:52,752 ALTERNATIVE HYPOTHESIS 112 00:05:52,819 --> 00:05:54,053 REPRESENTS WHAT THE INVESTIGATOR 113 00:05:54,120 --> 00:05:58,624 AIMS TO SHOW. FOR EXAMPLE, A 114 00:05:58,691 --> 00:06:01,260 HYPOTHESIS OF THERE IS NO 115 00:06:01,327 --> 00:06:03,096 DIFFERENCE BETWEEN THE TWO 116 00:06:03,162 --> 00:06:13,706 TREATMENTS IS NULL HYPOTHESIS. 117 00:06:15,475 --> 00:06:17,677 THE DIFFERENCE BETWEEN THE TWO 118 00:06:17,744 --> 00:06:19,779 DIFFERENT TREATMENTS IS A NULL 119 00:06:19,846 --> 00:06:22,515 HYPOTHESIS. ONE TYPE IS CALLED 120 00:06:22,582 --> 00:06:26,085 AN ALPHAER ROR. IF WE REJECT 121 00:06:26,152 --> 00:06:28,955 THE NULL HYPOTHESIS, WHETHER IT 122 00:06:29,021 --> 00:06:31,090 IS TRUE, WE COMMIT TYPE ONE 123 00:06:31,157 --> 00:06:35,361 ERROR. FOR EXAMPLE, WHEN 124 00:06:35,428 --> 00:06:39,866 THERE'S NO DIFFERENCE HERE WE 125 00:06:39,932 --> 00:06:40,933 FALSELY CONCLUDE THERE IS A 126 00:06:41,000 --> 00:06:43,903 DIFFERENCE. THE OTHER TYPE OF 127 00:06:43,970 --> 00:06:46,205 ERROR IS CALLED BETA ERROR IF WE 128 00:06:46,272 --> 00:06:48,875 FAIL TO REJECT THE NULL 129 00:06:48,941 --> 00:06:52,145 HYPOTHESIS AND THE NULL 130 00:06:52,211 --> 00:06:54,347 HYPOTHESIS IS NOT TRUE WE COMMIT 131 00:06:54,413 --> 00:06:56,315 ABETA ERROR. FOR EXAMPLE, WE -- 132 00:06:56,382 --> 00:06:58,217 THERE'S A DIFFERENCE BETWEEN THE 133 00:06:58,284 --> 00:07:01,487 TWO TREATMENTS. 134 00:07:01,554 --> 00:07:06,292 WE FALSELY CONCLUDE THAT THERE'S 135 00:07:06,359 --> 00:07:10,229 NO DIFFERENCE. THE POWER IS 136 00:07:10,296 --> 00:07:19,372 BASED ON TYPE TWO ERROR. THE 137 00:07:19,438 --> 00:07:26,712 POWER IS DEFINED AS NOT A TWO 138 00:07:26,779 --> 00:07:28,381 ERROR. THIS IS DEFINED AS A 139 00:07:28,447 --> 00:07:31,250 PROBABILITY OF CONCLUDING THAT 140 00:07:31,317 --> 00:07:33,252 THERE IS A DIFFERENCE BETWEEN 141 00:07:33,319 --> 00:07:36,455 THE TWO TREATMENTS. WHEN THE 142 00:07:36,522 --> 00:07:38,858 DIFFERENCE ACTUALLY EXISTS, FOR 143 00:07:38,925 --> 00:07:43,729 EXAMPLE, A STUDY TO COMPARE A 144 00:07:43,796 --> 00:07:47,099 DRUG WITH. THE DRUG IS TRULY 145 00:07:47,166 --> 00:07:48,634 EFFECTIVE. THE NULL HYPOTHESIS 146 00:07:48,701 --> 00:07:54,740 IS A DIFFERENCE BETWEEN NEW ER 147 00:07:54,807 --> 00:07:58,244 WHERE IT IS THE DRUG TREATMENT 148 00:07:58,311 --> 00:08:00,079 AND THE CONTROL POPULATION. IF 149 00:08:00,146 --> 00:08:06,719 WE SPECIFY POWER EQUAL TO 0.8 OR 150 00:08:06,786 --> 00:08:11,157 TYPE TWO ERROR AS 0.2, IMAGINE 151 00:08:11,224 --> 00:08:15,995 THIS STUDY WILL BECOME CONDUCTED 152 00:08:16,062 --> 00:08:18,464 MANY TIMES. THIS MEANS THAT 80% 153 00:08:18,531 --> 00:08:21,467 OF THE TIMES WE WILL REJECT THAT 154 00:08:21,534 --> 00:08:24,303 HYPOTHESIS AND CONCLUDE THAT 155 00:08:24,370 --> 00:08:27,340 THERE IS A STATISTICAL 156 00:08:27,406 --> 00:08:29,375 SIGNIFICANT DIFFERENCE BETWEEN 157 00:08:29,442 --> 00:08:30,910 DRUG A AND THE SPECIFIC GROUPS 158 00:08:30,977 --> 00:08:34,213 HOWEVER THAT 20% OF THE TIMES WE 159 00:08:34,280 --> 00:08:38,951 WILL ACCEPT THE HYPOTHESIS AND 160 00:08:39,018 --> 00:08:44,056 CONCLUDE THAT THERE'S NO 161 00:08:44,123 --> 00:08:45,725 SIGNIFICANT DIFFERENCE BETWEEN 162 00:08:45,791 --> 00:08:47,193 THE TWO GROUPS EVEN THOUGH THE 163 00:08:47,260 --> 00:08:57,803 DRUG IS TRULY EFFECTIVE. SINCE 164 00:09:02,642 --> 00:09:05,645 THE POWER ANALYSIS IS DEPENDING 165 00:09:05,711 --> 00:09:07,346 ON THE TEST METHOD I WILL GIVE 166 00:09:07,413 --> 00:09:10,216 YOU AN OVERVIEW OF THREE 167 00:09:10,283 --> 00:09:18,424 COMMONLY USED DESIGNS. THE 168 00:09:18,491 --> 00:09:22,561 RANDOMIZED CONTROL TRIAL IS THE 169 00:09:22,628 --> 00:09:23,129 MOST IMPORTANT TRIAL IN THE 170 00:09:23,195 --> 00:09:27,366 CLINICAL TRIAL STUDY. IN THE 171 00:09:27,433 --> 00:09:30,436 CLINICAL STUDY. WHICH HAS THREE 172 00:09:30,503 --> 00:09:30,937 FEATURES, INTERVENTION, 173 00:09:31,003 --> 00:09:34,807 RANDOMIZATION AND CONTROL. THIS 174 00:09:34,874 --> 00:09:37,710 IS A GOLD STANDARD FOR 175 00:09:37,777 --> 00:09:39,345 EVALUATING THE EFFECTIVENESS OF 176 00:09:39,412 --> 00:09:43,849 EVALUATION. THE PARALLEL TWO 177 00:09:43,916 --> 00:09:48,587 GROUP DESIGN IS THE SIMPLEST 178 00:09:48,654 --> 00:09:50,222 RANDOMIZED CONTROL TRIAL. IT 179 00:09:50,289 --> 00:09:51,357 INCLUDES ONLY TWO GROUPS, 180 00:09:51,424 --> 00:09:54,427 CONTROL AND TREATMENT. SUBJECTS 181 00:09:54,493 --> 00:09:57,396 ARE RANDOMLY ASSIGNED TO THE 182 00:09:57,463 --> 00:09:59,598 CONTROL GROUP OR TREATMENT 183 00:09:59,665 --> 00:10:04,937 GROUP. T THE NULL HYPOTHESIS IS 184 00:10:05,004 --> 00:10:06,872 A DIFFERENCE BETWEEN QUARTER 185 00:10:06,939 --> 00:10:10,209 ZERO. TO TEST THE HYPOTHESIS, 186 00:10:10,276 --> 00:10:17,083 WE USE TWO SAMPLE T-TEST. WITH 187 00:10:17,149 --> 00:10:22,788 NO SO VARIETY. WITH COVARY YAT 188 00:10:22,855 --> 00:10:23,622 WE USE AN ANALYSIS. 189 00:10:23,689 --> 00:10:26,625 THE SECOND DESIGN IS CROSSOVER 190 00:10:26,692 --> 00:10:29,562 DESIGN. CROSSOVER DESIGN IS A 191 00:10:29,628 --> 00:10:31,364 SPECIFIC TYPE OF RANDOMIZED 192 00:10:31,430 --> 00:10:41,941 CONTROL TRIAL. THE CROSSOVER 193 00:10:44,076 --> 00:10:46,212 DESIGN, THE AB DESIGN IS THE 194 00:10:46,278 --> 00:10:49,849 SIMPLEST ONE. THE AB/BA 195 00:10:49,915 --> 00:10:52,418 INCLUDES TWO SEQUENCE, AB OR BA. 196 00:10:52,485 --> 00:10:55,187 THE SUBJECTS ASSIGNED TO THE AB 197 00:10:55,254 --> 00:10:56,422 SEQUENCE RECEIVE TREATMENT A IN 198 00:10:56,489 --> 00:10:59,358 THE FIRST PERIOD AND TREATMENT B 199 00:10:59,425 --> 00:11:02,828 IN THE SECOND PERIOD AND VICE 200 00:11:02,895 --> 00:11:05,264 VERSA. SUBJECTS ARE RANDOMLY 201 00:11:05,331 --> 00:11:09,602 ASSIGNED TO SEQUENCE AB OR BA. 202 00:11:09,668 --> 00:11:12,772 HERE, THE RANDOMIZATION IS 203 00:11:12,838 --> 00:11:16,242 DIFFERENT FROM THAT USE IN 204 00:11:16,308 --> 00:11:18,177 PARALLEL DESIGN. IN PARALLEL 205 00:11:18,244 --> 00:11:21,547 DESIGN SUBJECTS ARE RANDOMLY 206 00:11:21,614 --> 00:11:24,784 ASSIGNED TO A OR B AND EACH 207 00:11:24,850 --> 00:11:28,220 SUBJECT HAS THEIR OWN TREATMENT. 208 00:11:28,287 --> 00:11:29,388 AND THE SUBJECT RECEIVED TWO 209 00:11:29,455 --> 00:11:30,790 TREATMENT IN THE OTHER DESIGN. 210 00:11:30,856 --> 00:11:37,830 TWO TREATMENT IN DIFFERENT 211 00:11:37,897 --> 00:11:39,331 PERIOD. THEREFORE THEY SERVE AS 212 00:11:39,398 --> 00:11:41,467 HIS OR HER OWN CONTROL. THE KEY 213 00:11:41,534 --> 00:11:42,802 ASSUMPTION FOR CROSSOVER DESIGN 214 00:11:42,868 --> 00:11:46,105 IS NO CARRY OVEREFFECT. THIS 215 00:11:46,172 --> 00:11:47,606 MEANS THE TREATMENT ASSIGNED 216 00:11:47,673 --> 00:11:50,709 DURING THE FIRST PERIOD DOES NOT 217 00:11:50,776 --> 00:11:53,479 INTERFERE WITH THE OUTCOME IN 218 00:11:53,546 --> 00:12:01,987 THE SECOND PERIOD. SO NOW THE 219 00:12:02,054 --> 00:12:07,359 HYPOTHESIS IS 1 AND 0. WE USE 220 00:12:07,426 --> 00:12:11,063 THE PAIRED TEST T-TEST. AND ONE 221 00:12:11,130 --> 00:12:12,665 SAMPLE T-TEST IS BASED ON THE 222 00:12:12,731 --> 00:12:15,000 DIFFERENCE BETWEEN THE PAIRED 223 00:12:15,067 --> 00:12:16,235 OBSERVATIONS. WITH COVARIATES 224 00:12:16,302 --> 00:12:20,473 WE USE REPEATED MEASURES AN KOE 225 00:12:20,539 --> 00:12:26,846 HAVE A. 226 00:12:26,912 --> 00:12:31,383 THE SECOND DESIGN IS A QUASI 227 00:12:31,450 --> 00:12:34,653 EXPERIMENTAL DESIGN. THE 228 00:12:34,720 --> 00:12:39,358 FIRST -- THE FIRST IS CALLED THE 229 00:12:39,425 --> 00:12:41,660 PARALLEL DESIGN WHICH HAS THE 230 00:12:41,727 --> 00:12:46,799 THREE FEATURES. RANDOMIZATION, 231 00:12:46,866 --> 00:12:50,503 CONTROL AND INDIVIDUAL. FOR THE 232 00:12:50,569 --> 00:12:51,971 QUASI EXPERIMENTAL DESIGN THAT'S 233 00:12:52,037 --> 00:12:54,974 NOT -- QUASI EXPERIMENTAL DESIGN 234 00:12:55,040 --> 00:13:00,779 DOES NOT USE RANDOMIZATION. THE 235 00:13:00,846 --> 00:13:02,915 ADVANTAGE OF THE QUASI 236 00:13:02,982 --> 00:13:05,851 EXPERIMENTAL DESIGN IS LESS 237 00:13:05,918 --> 00:13:08,821 EXPENSIVE. THE WEAKNESS IS THE 238 00:13:08,888 --> 00:13:12,725 RESULTS ARE SUBJECT TO BIAS. 239 00:13:12,791 --> 00:13:16,829 THEREFORE, THESE DESIGNS ARE 240 00:13:16,896 --> 00:13:19,331 USED IN HIGH LEVEL STUDY OR 241 00:13:19,398 --> 00:13:22,334 EARLY PHASE STUDY. ONE GROUP 242 00:13:22,401 --> 00:13:28,207 PRETEST THE POSTTEST DESIGN IS 243 00:13:28,274 --> 00:13:31,744 SIMPLEST. QUASI EXPERIMENTAL 244 00:13:31,810 --> 00:13:38,217 DESIGN. THIS DESIGN EXCEPT THE 245 00:13:38,284 --> 00:13:41,320 RANDOMIZATION HAS NOT CONTROL 246 00:13:41,387 --> 00:13:42,855 GROUP. IT INCLUDE ONLY ONE 247 00:13:42,922 --> 00:13:47,359 GROUP. AND ALL SUBJECTS IN THE 248 00:13:47,426 --> 00:13:50,529 GROUP RECEIVE THE TREATMENT. 249 00:13:50,596 --> 00:13:52,231 BUT THEN THERE'S THE OUTCOME 250 00:13:52,298 --> 00:13:54,733 VARIABLES ARE COLLECTED BEFORE 251 00:13:54,800 --> 00:13:56,902 AND AFTER THE TREATMENT. THE 252 00:13:56,969 --> 00:13:59,338 EFFECT OF TREATMENT IS EVALUATED 253 00:13:59,405 --> 00:14:06,545 BY COMPARING THE POSTTEST WITH 254 00:14:06,612 --> 00:14:08,914 PRETEST. THEREFORE, EACH 255 00:14:08,981 --> 00:14:12,184 SUBJECT SERVES AS THEIR OWN 256 00:14:12,251 --> 00:14:14,587 CONTROL. THIS IS SIMILAR AS 257 00:14:14,653 --> 00:14:18,757 AB/BA CROSSOVER DESIGN. THE 258 00:14:18,824 --> 00:14:23,596 DIFFERENCE IS IN AB/BA CROSSOVER 259 00:14:23,662 --> 00:14:25,998 DESIGN, THERE ARE TWO SEQUENCES 260 00:14:26,065 --> 00:14:30,736 AB AND BA. AND EACH SUBJECT IS 261 00:14:30,803 --> 00:14:34,173 RANDOMIZED TO RECEIVE AB OR BA 262 00:14:34,240 --> 00:14:37,243 SECRETS. BUT IN THE ONE GROUP 263 00:14:37,309 --> 00:14:39,845 DESIGN THERE'S ONLY ONE 264 00:14:39,912 --> 00:14:44,250 SEQUENCE. CALLED AB AND ALL 265 00:14:44,316 --> 00:14:49,455 SUBJECTS RECEIVE SAME SEQUENCE. 266 00:14:49,521 --> 00:14:53,559 BUT NOW HYPOTHESIS IS HERE AND 267 00:14:53,626 --> 00:14:56,262 TO TEST THE HYPOTHESIS WE ALSO 268 00:14:56,328 --> 00:14:59,098 USE PAIRED T-TEST WHICH 269 00:14:59,164 --> 00:15:00,666 EQUIVALENT TO THE T-TEST BASED 270 00:15:00,733 --> 00:15:04,136 ON THE DIFFERENCE BETWEEN THE 271 00:15:04,203 --> 00:15:07,740 PAIRED OBSERVATION. NEXT, I AM 272 00:15:07,806 --> 00:15:08,974 GOING TO PRESENT HOW TO 273 00:15:09,041 --> 00:15:11,343 CALCULATE THE SAMPLE SIZE FOR 274 00:15:11,410 --> 00:15:19,351 TWO SAMPLE T-TEST AND PAIRED 275 00:15:19,418 --> 00:15:23,922 T-TEST. FOR THE PARALLEL 276 00:15:23,989 --> 00:15:27,159 DESIGN -- TWO GROUP PARALLEL 277 00:15:27,226 --> 00:15:31,397 DESIGN THE HYPOTHESIS IS THE 278 00:15:31,463 --> 00:15:35,067 DIFFERENCE BETWEEN THE TWO AND 279 00:15:35,134 --> 00:15:37,970 ZERO. NOTICE THAT THE DATA IS 280 00:15:38,037 --> 00:15:40,139 NOTED AS THE DIFFERENCE THAT HAS 281 00:15:40,205 --> 00:15:40,639 SCIENTIFIC OR CLINICAL 282 00:15:40,706 --> 00:15:46,745 IMPORTANCE. SIGMA IS DENOTED AS 283 00:15:46,812 --> 00:15:48,814 THE POPULAR STANDARD OF 284 00:15:48,881 --> 00:15:52,217 DEVIATION. THE DATA TO SIGMA IS 285 00:15:52,284 --> 00:15:55,354 STANDARDIZED MEAN DIFFERENCE OR 286 00:15:55,421 --> 00:15:57,523 STANDARDIZED EFFECT SIZE AND A 287 00:15:57,589 --> 00:15:59,958 TYPE ONE AND TYPE TWO ERRORS. 288 00:16:00,025 --> 00:16:04,229 THE METHOD FOR THE SAMPLE SIZE 289 00:16:04,296 --> 00:16:08,701 ESTIMATION IS WOE USE THE TWO 290 00:16:08,767 --> 00:16:11,170 SAMPLE TWO SAILED Z TEST. THE 291 00:16:11,236 --> 00:16:14,606 FORMULA TO CALCULATE THE SAMPLE 292 00:16:14,673 --> 00:16:25,217 SIZE IS THIS. AS THIS HERE. 293 00:16:33,692 --> 00:16:40,232 IS AS THIS AND HERE THE Z AT 294 00:16:40,299 --> 00:16:44,370 ALPHA AND Z AT BETA ARE THE 295 00:16:44,436 --> 00:16:45,204 CRITICAL VALUES OF THE 296 00:16:45,270 --> 00:16:45,804 STANDARDIZED NORMAL 297 00:16:45,871 --> 00:16:48,340 DISTRIBUTION. HERE WE USE THE Z 298 00:16:48,407 --> 00:16:50,776 TEST INSTEAD OF THE T-TEST 299 00:16:50,843 --> 00:16:53,946 BECAUSE THEY USE THE POPULATION 300 00:16:54,012 --> 00:16:54,646 TARRED DEVIATION AND NOT USE THE 301 00:16:54,713 --> 00:16:58,250 DEGREE OF FREEDOM. HERE WE CAN 302 00:16:58,317 --> 00:16:59,351 HAVE A SIMPLE FORMULA. 303 00:16:59,418 --> 00:17:03,355 THE T-TEST THAT USES THE SAMPLE 304 00:17:03,422 --> 00:17:07,359 DEVIATION AND THE FREEDOM 305 00:17:07,426 --> 00:17:09,695 INDEED, THE TWO METHODS GIVE 306 00:17:09,762 --> 00:17:15,367 SIMILAR SAMPLE SIZE WHEN SAMPLE 307 00:17:15,434 --> 00:17:22,241 SIZE GREATER THAN 50. THOSE 308 00:17:22,307 --> 00:17:23,642 SLIDES SHOW THE SAMPLE SIZE FOR 309 00:17:23,709 --> 00:17:31,350 PAIRED T-TEST. THE DATA OF THE 310 00:17:31,417 --> 00:17:33,485 DEFINITION SAME AS BEFORE. AND 311 00:17:33,552 --> 00:17:36,688 HERE, THE DIFFERENCE IS SIGMA D. 312 00:17:36,755 --> 00:17:41,960 SIGMA D IS THE POPULAR STANDARD 313 00:17:42,027 --> 00:17:43,362 DEVIATION OF THE PAIRED 314 00:17:43,429 --> 00:17:50,502 DIFFERENCE. IF THE SIGMA NOTED 315 00:17:50,569 --> 00:17:51,303 AS THE POPULATION STANDARD 316 00:17:51,370 --> 00:17:54,306 DEVIATION AND WE CAN USE THIS 317 00:17:54,373 --> 00:17:58,877 FORMULA TO CALCULATE THE P 318 00:17:58,944 --> 00:18:00,245 POPULATION DEVIATION OF THE 319 00:18:00,312 --> 00:18:02,281 PAIRED DIFFERENCE HERE BELOW IS 320 00:18:02,347 --> 00:18:03,348 THE CORRELATION COEFFICIENT 321 00:18:03,415 --> 00:18:06,151 BETWEEN THE PAIRED OBSERVATIONS. 322 00:18:06,218 --> 00:18:10,656 BASED ON THIS FORMULA, WHEN IT 323 00:18:10,722 --> 00:18:12,324 IS 0.5 THE SIGMA D IS EQUAL TO 324 00:18:12,391 --> 00:18:22,935 SIGMA. AND THEN WE HAVE SIGMA D 325 00:18:24,736 --> 00:18:28,240 IS GREATER THAN SIGMA. SO WE 326 00:18:28,307 --> 00:18:34,213 DON'T IN THE POWER ANALYSIS, WE 327 00:18:34,279 --> 00:18:37,282 DON'T KNOW THIS. USUALLY THIS 328 00:18:37,349 --> 00:18:38,617 IS EQUAL TO ZERO OR THE FIVE 329 00:18:38,684 --> 00:18:43,822 WHICH MEANS WE ASSUME THE 330 00:18:43,889 --> 00:18:46,024 POPULA 331 00:18:46,091 --> 00:18:49,094 POPULATION STANDARD DEVIATION IS 332 00:18:49,161 --> 00:18:51,296 EQUAL TO THE POPULATION STANDARD 333 00:18:51,363 --> 00:18:52,231 DEVIATION AND HERE IS THE 334 00:18:52,297 --> 00:18:54,233 EXAMPLE WHERE WE USE THE TWO 335 00:18:54,299 --> 00:18:57,102 TAILED Z TEST AND WE'RE WE GET 336 00:18:57,169 --> 00:19:03,208 THIS FORMULA TO THE FORMULA OF 337 00:19:03,275 --> 00:19:05,944 THE EXAMPLE T-TEST. YOU SEE THE 338 00:19:06,011 --> 00:19:07,579 ONLY DIFFERENCE IN THIS FORMULA 339 00:19:07,646 --> 00:19:13,552 IS TWO. THEREFORE, WHEN SIGMA D 340 00:19:13,619 --> 00:19:16,221 EQUAL TO SIGMA AND THE DATA AND 341 00:19:16,288 --> 00:19:21,226 ALPHA AND THE BETA ARE KEPT THE 342 00:19:21,293 --> 00:19:24,229 SAME, THEN THE SAME SIZE FOR THE 343 00:19:24,296 --> 00:19:28,233 PARALLEL DESIGN WILL HAVE THE 344 00:19:28,300 --> 00:19:30,736 SAMPLE DESIGN IS FOUR TIMES OF 345 00:19:30,802 --> 00:19:34,206 THE SAMPLE SIZE FOR THE 346 00:19:34,273 --> 00:19:38,777 CROSSOVER DESIGN. THIS FORMULA, 347 00:19:38,844 --> 00:19:41,613 WE -- WHEN WE DO PAIRED ANALYSIS 348 00:19:41,680 --> 00:19:48,620 WE NEED ALPHA, BETA, THETA AND 349 00:19:48,687 --> 00:19:59,197 SIGMA. NEXT SLIDE SH, . ALPHA 350 00:19:59,264 --> 00:20:00,365 THE PROBABILITY OF MAKING TYPE 351 00:20:00,432 --> 00:20:03,068 ONE ERROR AND THE BETA IS THE 352 00:20:03,135 --> 00:20:05,070 PROBABILITY OF MAKING TYPE TWO 353 00:20:05,137 --> 00:20:06,972 ERROR. IDEAL THAT WE WANT TO 354 00:20:07,039 --> 00:20:09,508 MINIMIZE THE PROBABILITY OF 355 00:20:09,575 --> 00:20:14,012 COMMITTING ALPHA AND BETA ERROR. 356 00:20:14,079 --> 00:20:15,180 FOR CONSIDERATION OF SAMPLE SIZE 357 00:20:15,247 --> 00:20:19,685 WE USUALLY SAID BETA EQUAL TO 358 00:20:19,751 --> 00:20:22,754 0.2. AND ALPHA EQUAL TO 0.05. 359 00:20:22,821 --> 00:20:27,292 YOU MAY WONDER, WHY WE CHOOSING 360 00:20:27,359 --> 00:20:31,229 SMALLER TYPE ONE ERROR THAN TYPE 361 00:20:31,296 --> 00:20:32,464 TWO ERROR. THE MAIN REASON IS 362 00:20:32,531 --> 00:20:36,101 THAT IN MOST SITUATIONS MAKING 363 00:20:36,168 --> 00:20:38,337 THAT TYPE ONE ERROR MIGHT BE 364 00:20:38,403 --> 00:20:40,205 MORE SEVERE OR COSTLY THAN 365 00:20:40,272 --> 00:20:45,510 MAKING TYPE TWO ER ERROR THIS 366 00:20:45,577 --> 00:20:48,747 BALANCE IS CONSIDERED REASONABLE 367 00:20:48,814 --> 00:20:53,518 IN MANY STUDIES. FOR EXAMPLE, 368 00:20:53,585 --> 00:20:54,620 FIRSTLY DECLARING A DRUG TO BE 369 00:20:54,686 --> 00:20:56,855 EFFECTIVE WHEN IT'S NOT. 370 00:20:56,922 --> 00:21:00,926 COULD LEAD TO HARMFUL 371 00:21:00,993 --> 00:21:02,494 CONSEQUENCES FOR PATIENTS. 372 00:21:02,561 --> 00:21:04,796 THEREFORE, RESEARCHERS LIKE TO 373 00:21:04,863 --> 00:21:07,366 SET A HIGHER SIGNIFICANCE LABEL 374 00:21:07,432 --> 00:21:10,736 TO MINIMIZE THIS RISK SUCH AS 375 00:21:10,802 --> 00:21:21,313 0.01 OR 0.001. BASED ON THE 376 00:21:21,813 --> 00:21:26,885 FORMULA TO CORRECT -- 377 00:21:26,952 --> 00:21:29,021 CALCULATING SAMPLE SIZE WE NEED 378 00:21:29,087 --> 00:21:30,922 SIGMA, THE POPULATION STANDARD 379 00:21:30,989 --> 00:21:31,223 DEVIATION. 380 00:21:31,289 --> 00:21:37,829 AND THE DATHETA, THE SMALLEST 381 00:21:37,896 --> 00:21:39,364 DIFFERENCE BETWEEN TW TWO GROUPS 382 00:21:39,431 --> 00:21:42,734 THAT HAS SCIENTIFIC OR CLINICAL 383 00:21:42,801 --> 00:21:44,436 IMPORTANCE. TO GET TO THIS 384 00:21:44,503 --> 00:21:47,339 ESTIMATION FOR THIS TWO 385 00:21:47,406 --> 00:21:50,308 COMPONENTS, WE USUALLY GET THIS 386 00:21:50,375 --> 00:21:54,146 FROM LITERATURE REVIEW, PREVIOUS 387 00:21:54,212 --> 00:21:56,381 STUDIES, OR PILOT STUDY. IF 388 00:21:56,448 --> 00:21:59,518 THIS INFORMATION IS NOT 389 00:21:59,584 --> 00:22:07,359 AVAILABLE, WE CAN USE COHEN'S 390 00:22:07,426 --> 00:22:10,228 RECOMMENDATION. COHEN'S 391 00:22:10,295 --> 00:22:14,700 RECOMMENDATION IS BASED ON 392 00:22:14,766 --> 00:22:19,071 COHEN'S D. COHEN'S D IS DEFINED 393 00:22:19,137 --> 00:22:20,238 AS THE DIFFERENCE BETWEEN GROUP 394 00:22:20,305 --> 00:22:23,141 ONE AND GROUP TWO MEANS DIVIDED 395 00:22:23,208 --> 00:22:26,278 BY THE POOLED STANDARD 396 00:22:26,344 --> 00:22:31,383 DEVIATION. COHEN DEFINED D 397 00:22:31,450 --> 00:22:35,454 EQUALS .25 AS TOO SMALL TO 398 00:22:35,520 --> 00:22:45,964 DEFECT AND MEDIUM AND .5 399 00:22:47,199 --> 00:22:53,038 CAREFUL OBSERVATION. COHEN 400 00:22:53,105 --> 00:22:56,007 RECOMMEND THAT 0.5 CAN BE 401 00:22:56,074 --> 00:22:57,676 CLINICAL IMPORTANT EFFECT. YOU 402 00:22:57,743 --> 00:22:59,344 CAN EASILY CALCULATE THE SAMPLE 403 00:22:59,411 --> 00:23:09,921 SIZE BASED ON COHEN'S D EQUAL 404 00:23:10,355 --> 00:23:12,290 0.5. SO IN THE FORMULA, WE CAN 405 00:23:12,357 --> 00:23:19,464 REPLACE THE 0.5 TO THE DATA TO 406 00:23:19,531 --> 00:23:22,234 SIGMA. THE EFFECTIVE SIZE AND 407 00:23:22,300 --> 00:23:32,844 FOR ALPHA EQUAL 0.05 AND FOR 0.8 408 00:23:35,914 --> 00:23:43,054 WE CAN GET Z HALF ALPHA EQUAL TO 409 00:23:43,121 --> 00:23:43,789 1.96. REPLACE THIS THREE NUMBER 410 00:23:43,855 --> 00:23:47,959 TO THE FORMULA. WE GET 63. AND 411 00:23:48,026 --> 00:23:51,663 FOR THE -- IF WE USE A T-TEST. 412 00:23:51,730 --> 00:23:58,203 WE GET 64. LIP SI AND WILSON 413 00:23:58,270 --> 00:24:02,507 DID A META-ANALYSIS OF 302 414 00:24:02,574 --> 00:24:04,342 METAANALYSIS OF OVER 10,000 415 00:24:04,409 --> 00:24:08,280 STUDIES AND FOUND THAT THE 416 00:24:08,346 --> 00:24:11,917 AVERAGE EFFECT SIZE WAS 0.5 417 00:24:11,983 --> 00:24:19,357 WHICH IS PART OF COHEN'S 418 00:24:19,424 --> 00:24:24,729 RECOMMENDATION. KING RECOGNIZED 419 00:24:24,796 --> 00:24:30,435 THAT A STANDARD EFFECT SIZE OF 420 00:24:30,502 --> 00:24:37,776 0.3 CAN BE APPEAR EFFECT -- AN 421 00:24:37,843 --> 00:24:38,677 EFFECTIVE EFFECT. 422 00:24:38,743 --> 00:24:40,846 A PILOT STUDY HAS A SMALL SAMPLE 423 00:24:40,912 --> 00:24:44,983 SIZE SO WHEN USING THE SAMPLE 424 00:24:45,050 --> 00:24:47,352 STANDARD DEVIATION FROM PILOT 425 00:24:47,419 --> 00:24:50,689 STUDY WE SHOULD BE WEARY THE 426 00:24:50,755 --> 00:24:51,990 SAMPLE STANDARD DEVIATION CAN BE 427 00:24:52,057 --> 00:24:55,360 EITHER LARGER OR SMALLER THAN 428 00:24:55,427 --> 00:24:58,230 THAT OF THE MAIN STUDY. THE 429 00:24:58,296 --> 00:25:00,699 DISTRIBUTION OF THE SAMPLE 430 00:25:00,765 --> 00:25:06,972 VARIANCE IS POSITIVELY SKEWED 431 00:25:07,038 --> 00:25:08,707 WITH THE LONG RIGHT TAIL WHICH 432 00:25:08,773 --> 00:25:14,246 MEANS THAT OVER 50% OF THE TIME, 433 00:25:14,312 --> 00:25:24,456 A REALLY LONG VARIANCE -- USING 434 00:25:24,522 --> 00:25:27,359 AN UNDERESTIMATE OF STANDARD OF 435 00:25:27,425 --> 00:25:29,327 DEVIATION WILL YIELD AN 436 00:25:29,394 --> 00:25:30,862 UNDERESTIMATE OF SAMPLE SIZE AND 437 00:25:30,929 --> 00:25:35,367 WILL RESULT IN LESS POWER THAN 438 00:25:35,433 --> 00:25:37,168 ANTICIPATED. BELONG THESE ARE 439 00:25:37,235 --> 00:25:41,006 THE TWO TAILED TWO SAMPLE T-TEST 440 00:25:41,072 --> 00:25:44,542 SHOW THAT HAD. FOR TRIALS WITH 441 00:25:44,609 --> 00:25:45,510 SAMPLE SIZE ESTIMATED FROM 442 00:25:45,577 --> 00:25:50,015 SAMPLE STANDARD DEVIATION UPON 443 00:25:50,081 --> 00:25:53,985 ESTIMATE OF POPULATION STANDARD 444 00:25:54,052 --> 00:25:54,653 DEVIATION. THE PROBABILITY THAT 445 00:25:54,719 --> 00:25:57,055 THE TRIALS ACHIEVE THE POWER 446 00:25:57,122 --> 00:26:01,192 OVER 0.8 IS ONLY ABOUT 50%. FOR 447 00:26:01,259 --> 00:26:04,229 THE TRIALS, WITH THE SAMPLE SIZE 448 00:26:04,296 --> 00:26:08,733 ESTIMATED FROM THE 80% UP ONE 449 00:26:08,800 --> 00:26:11,202 SIDED CONFIDENCE LIMIT ON THE 450 00:26:11,269 --> 00:26:12,737 POPULATION STANDARD OF DEVIATION 451 00:26:12,804 --> 00:26:15,373 THE PROBABILITY THAT THE TRIALS 452 00:26:15,440 --> 00:26:19,344 ACHIEVE THE POWER GREATER THAT 453 00:26:19,411 --> 00:26:27,352 0.8 IS OVER 80%. FOR EXAMPLE, 454 00:26:27,419 --> 00:26:34,526 FROM THE PILOT STUDY WE CAN 455 00:26:34,592 --> 00:26:37,295 ESTIMATE THAT WE CAN SEE THERE'S 456 00:26:37,362 --> 00:26:39,331 60% CONFIDENCE AS THIS. AND 457 00:26:39,397 --> 00:26:44,302 THEN WE WILL GET TO THE 80% UP 458 00:26:44,369 --> 00:26:49,374 ONE SIDED COST LIMITED AS 2.76 459 00:26:49,441 --> 00:26:51,876 THIS MEANS WE ARE 80% SURE THAT 460 00:26:51,943 --> 00:26:54,546 THE TRUE VALUE OF POPULATION 461 00:26:54,612 --> 00:26:56,948 STANDARD OF DEVIATION IS LESS OR 462 00:26:57,015 --> 00:27:07,492 EQUAL TO 2.76. NEXT WE ARE 463 00:27:08,360 --> 00:27:11,363 GOING TO DISCUSS THE FACTORS 464 00:27:11,429 --> 00:27:16,201 EFFECTING THE STATISTICAL POWER. 465 00:27:16,267 --> 00:27:19,437 LET'S LOOK AT THESE FORMULA 466 00:27:19,504 --> 00:27:23,341 AGAIN. THE LARGEST SAMPLE SIZE 467 00:27:23,408 --> 00:27:27,178 CAN BE CAUSED BY LARGE STANDARD 468 00:27:27,245 --> 00:27:31,783 OF DEVIATION. LARGE Z AT HALF 469 00:27:31,850 --> 00:27:37,956 ALPHA OR LARGE AT BETA. Z BETA. 470 00:27:38,023 --> 00:27:43,094 OR SMALL DATA. THE LARGE SIGMA 471 00:27:43,161 --> 00:27:45,964 MEANS LARGE POPULATION VARIATION 472 00:27:46,031 --> 00:27:48,099 OF OUTCOME. FOR EXAMPLE, THE 473 00:27:48,166 --> 00:27:52,203 SAMPLE SIZE CALCULATED BASED ON 474 00:27:52,270 --> 00:27:55,640 SIGMA TWO WILL BE LARGER THAN Z 475 00:27:55,707 --> 00:28:01,379 FROM SIGMA IF ONE. THE LARGEST 476 00:28:01,446 --> 00:28:05,984 Z AT HALF ALPHA MEANS SMALL TYPE 477 00:28:06,051 --> 00:28:09,821 ONE ERROR. FOR EXAMPLE, THE 478 00:28:09,888 --> 00:28:13,058 SAMPLE SIZE CALCULATED BASED ON 479 00:28:13,124 --> 00:28:16,494 ALPHA WAS 0.01 IS LARGER THAN 480 00:28:16,561 --> 00:28:22,834 THAT FROM 0.05. LARGE Z BETA 481 00:28:22,901 --> 00:28:26,938 MEANS SMALL TYPE TWO ERROR. OR 482 00:28:27,005 --> 00:28:31,910 LARGE POWER. THE SAMPLE SIZE 483 00:28:31,976 --> 00:28:35,080 CALCULATED BASE, FOR EXAMPLE, ON 484 00:28:35,146 --> 00:28:39,184 THE POWER OF 0.9 IS LARGER THAT 485 00:28:39,250 --> 00:28:44,789 THAT BASED ON POWER OF 0.8. THE 486 00:28:44,856 --> 00:28:47,358 SMALL DATA MEANS SMALL TREATMENT 487 00:28:47,425 --> 00:28:49,461 EFFECT. FOR EXAMPLE, THE SAMPLE 488 00:28:49,527 --> 00:28:51,830 SIZE CALCULATED BASED ON DATA 489 00:28:51,896 --> 00:28:56,301 EQUAL TO ONE IS LARGER THAN THAT 490 00:28:56,367 --> 00:28:58,203 BASED ON DATA, BASED ON THAT 491 00:28:58,269 --> 00:29:03,241 FROM DATA EQUAL TO 0. -- EQUAL 492 00:29:03,308 --> 00:29:07,745 TO 2. THIS TABLE SHOWS 493 00:29:07,812 --> 00:29:09,881 RELATIONSHIP BETWEEN SAMPLE SIZE 494 00:29:09,948 --> 00:29:14,219 TYPE ONE ERROR, STANDARDIZED 495 00:29:14,285 --> 00:29:17,388 EFFECT SIZE AND POWER. BY 496 00:29:17,455 --> 00:29:19,357 FIXING THE TYPE ONE ERROR AND 497 00:29:19,424 --> 00:29:21,159 THE STANDARDIZED EFFECT SIZE, 498 00:29:21,226 --> 00:29:25,864 DECREASING THE SAMPLE SIZE, 499 00:29:25,930 --> 00:29:30,802 DECREASING THE POWER. BY FIXING 500 00:29:30,869 --> 00:29:35,140 THE SAMPLE SIZE, AND THE 501 00:29:35,206 --> 00:29:36,341 STANDARDIZED EFFECT SIZE 502 00:29:36,407 --> 00:29:41,346 DECREASING THE TYPE ONE ERROR, 503 00:29:41,412 --> 00:29:45,483 DECREASING THE POWER. BY FIXING 504 00:29:45,550 --> 00:29:47,352 THE SAMPLE SIZE, AND THE TYPE 505 00:29:47,418 --> 00:29:50,188 ONE ERROR, DECREASING EFFECT 506 00:29:50,255 --> 00:29:56,394 SIZE, DECREASING THE POWER. 507 00:29:56,461 --> 00:29:59,364 THEREFORE THE FACTORS LEADING TO 508 00:29:59,430 --> 00:30:03,401 THE UNDERESTIMATED SAMPLE SIZE 509 00:30:03,468 --> 00:30:06,538 OF POWER IS UNDERESTIMATED 510 00:30:06,604 --> 00:30:10,708 STANDARD OF DEVIATION 511 00:30:10,775 --> 00:30:13,478 OVERESTIMATED EFFECTIVE SIZE OF 512 00:30:13,545 --> 00:30:15,346 LARGE TYPE ONE ERROR WHICH MEANS 513 00:30:15,413 --> 00:30:19,083 WE FAIL TO ACCOUNT FOR THE 514 00:30:19,150 --> 00:30:20,718 MULTIPLE COMPARISON. WHICH WILL 515 00:30:20,785 --> 00:30:21,586 BE PRESENT IN THE FOLLOWING 516 00:30:21,653 --> 00:30:32,197 SLIDES. THE PREVIOUS SLIDES ARE 517 00:30:32,964 --> 00:30:36,234 REALLY SHOW THE POWER -- THE 518 00:30:36,301 --> 00:30:40,238 TYPE TWO GO ERROR SLIDE 519 00:30:40,305 --> 00:30:43,408 DEVIATION EFFECT THE SAMPLE 520 00:30:43,474 --> 00:30:49,847 SIZE. THE FOLLOWING SLIDES I 521 00:30:49,914 --> 00:30:55,353 WILL PRESENT HOW STUDY DESIGN 522 00:30:55,420 --> 00:30:59,357 MULTIPLICITY AND THE GROUP 523 00:30:59,424 --> 00:31:01,993 WEIGHTS EFFECT THE ANALYSIS. 524 00:31:02,060 --> 00:31:04,229 THE FACTORS ARE OUTCOME VARIABLE 525 00:31:04,295 --> 00:31:07,165 TYPE, TRIAL TYPE, STATISTICAL 526 00:31:07,232 --> 00:31:09,767 ANALYSIS METHOD, DROPOUT RATE 527 00:31:09,834 --> 00:31:14,305 AND ET CETERA. WHICH I WILL NOT 528 00:31:14,372 --> 00:31:22,213 DISCUSS IN THIS PRESENTATION. 529 00:31:22,280 --> 00:31:23,448 IN THIS SLIDE SHOWS HOW THE 530 00:31:23,514 --> 00:31:25,717 EFFECT OF THE STUDY DESIGN 531 00:31:25,783 --> 00:31:27,018 EFFECTS THE SAMPLE SIZE. IN 532 00:31:27,085 --> 00:31:31,389 THIS TABLE, THE SAMPLE SIZE, THE 533 00:31:31,456 --> 00:31:33,691 TOTAL SAMPLE SIZE IS CALCULATED 534 00:31:33,758 --> 00:31:37,729 USING THE POWER OF ON EIGHT TYPE 535 00:31:37,795 --> 00:31:41,566 ONE ERROR 0.5 AND THE 536 00:31:41,633 --> 00:31:44,102 STANDARDIZED EFFECT SIDES OF 1.5 537 00:31:44,168 --> 00:31:46,204 AND TWO TAILED TEST. FOR 538 00:31:46,271 --> 00:31:50,742 PARALLEL GROUP DESIGN WE USE TWO 539 00:31:50,808 --> 00:31:52,343 SAMPLE T-TEST. BECAUSE EACH 540 00:31:52,410 --> 00:31:53,745 SUBJECT HAS ONLY ONE 541 00:31:53,811 --> 00:31:55,346 OBSERVATION. AND ALL THE 542 00:31:55,413 --> 00:31:59,417 SUBJECTS ARE INDEPENDENT. THE 543 00:31:59,484 --> 00:32:01,986 SAMPLE SIZE WERE 128. FOR THE 544 00:32:02,053 --> 00:32:04,222 CROSSOVER DESIGN AND ONE GROUP 545 00:32:04,289 --> 00:32:07,158 DESIGN WE USE PAIRED T-TEST 546 00:32:07,225 --> 00:32:10,228 BECAUSE EACH SUBJECT HAS TWO 547 00:32:10,295 --> 00:32:15,366 OBSERVATIONS AND TWO OKAY 548 00:32:15,433 --> 00:32:17,101 OBSERVATIONS ARE CORRELATED. 549 00:32:17,168 --> 00:32:20,371 AND IT WAS CORRELATED FROM THE 550 00:32:20,438 --> 00:32:21,973 OBSERVATION FROM THE SAME 551 00:32:22,040 --> 00:32:26,110 SUBJECT OF 0.3 WE CAN SAMPLE 552 00:32:26,177 --> 00:32:32,183 SIZE46 OF 0.5 AND WE HAVE 34. 553 00:32:32,250 --> 00:32:39,257 AND THE -- WE RECALL THE FORMULA 554 00:32:39,324 --> 00:32:42,193 FOR THE -- FROM THE T-TEST UP 555 00:32:42,260 --> 00:32:46,197 HERE THE T-TEST, THE DIFFERENCE 556 00:32:46,264 --> 00:32:48,800 IS THE -- IS A CONSTANT TWO. 557 00:32:48,866 --> 00:32:52,704 AND THE SAMPLE SIZE FOR THE 558 00:32:52,770 --> 00:32:55,073 PARALLEL DESIGN IS FOUR TIMES OF 559 00:32:55,139 --> 00:32:57,575 THE CROSSOVER DESIGN FOR THE 560 00:32:57,642 --> 00:32:59,344 STANDARD DEVIATION. OF THE 561 00:32:59,410 --> 00:33:02,413 DIFFERENCE IS EQUAL TO THE 562 00:33:02,480 --> 00:33:03,348 POPULATION STANDARD DEVIATION. 563 00:33:03,414 --> 00:33:06,617 SO HERE WE HAVE TO STUDY FOR IS 564 00:33:06,684 --> 00:33:13,291 ABOUT ONE QUARTER OF THE 128. 565 00:33:13,358 --> 00:33:15,126 IT IS OBVIOUS THAT IT IS A 566 00:33:15,193 --> 00:33:18,262 CROSSOVER DESIGN IS MUCH SMALLER 567 00:33:18,329 --> 00:33:21,666 SAMPLE SIZE COMPARED TO PAIRED 568 00:33:21,733 --> 00:33:22,900 GROUP DESIGN HOWEVER THE 569 00:33:22,967 --> 00:33:27,338 CROSSOVER DESIGN IS BASED ON 570 00:33:27,405 --> 00:33:31,409 STRONG AND STRICT ASSUMPTION 571 00:33:31,476 --> 00:33:36,514 WHICH IS NO CARRY OVER EFFECT. 572 00:33:36,581 --> 00:33:39,384 FOR THIS, FOR THIS ASSUMPTION, 573 00:33:39,450 --> 00:33:43,888 VERY FEW TREATMENTS CAN MEET 574 00:33:43,955 --> 00:33:44,889 THIS -- CAN MEET THIS ASSUM 575 00:33:44,956 --> 00:33:48,960 ASSUMPTION. THE OTHER 576 00:33:49,026 --> 00:33:50,027 ADVANTAGE -- DISADVANTAGE FOR 577 00:33:50,094 --> 00:33:57,268 THE CROSSOVER DESIGN IS A 578 00:33:57,335 --> 00:34:01,406 DROPOUT HAS A MORE NEGATIVE 579 00:34:01,472 --> 00:34:03,875 EFFECT THAN ON THE PAIRED 580 00:34:03,941 --> 00:34:07,278 DESIGN. FOR THE PARALLEL 581 00:34:07,345 --> 00:34:10,748 DESIGN, THE DESIGN CAN BE 582 00:34:10,815 --> 00:34:13,684 BALANCED OR UNBALANCED DEPENDS 583 00:34:13,751 --> 00:34:16,821 ON THE GROUP WEIGHTS. THE GROUP 584 00:34:16,888 --> 00:34:18,923 WEIGHTS ARE THE SAMPLE SIZE 585 00:34:18,990 --> 00:34:22,226 ALLOCATION WEIGHTS FOR THE TWO 586 00:34:22,293 --> 00:34:25,263 GROUPS. IF THE DESIGN HAS EQUAL 587 00:34:25,329 --> 00:34:27,365 WEIGHTS WE CAN BALANCE DESIGN IF 588 00:34:27,432 --> 00:34:30,334 THE DESIGN HAS UNEQUAL WEIGHTS 589 00:34:30,401 --> 00:34:35,072 WE CAN UNBALANCE THE DESIGN. IN 590 00:34:35,139 --> 00:34:40,211 THIS TABLE, THE SAMPLE SIZE IS 591 00:34:40,278 --> 00:34:47,118 CALCULATED BASED ON POWER OF 592 00:34:47,185 --> 00:34:51,489 0.8, ALPHA TYPE TWO AS 0.05. 593 00:34:51,556 --> 00:34:53,124 STANDARD DEVIATION OF ONE WHICH 594 00:34:53,191 --> 00:34:55,426 MEANS THE MEAN DIFFERENCE IS 595 00:34:55,493 --> 00:34:58,229 EQUIVALENT TO COHEN'S D AND WE 596 00:34:58,296 --> 00:35:04,235 USE THE TWO-TAILED T-TEST. FOR 597 00:35:04,302 --> 00:35:07,839 THE BALANCED DESIGN FOR THE 598 00:35:07,905 --> 00:35:12,210 EFFECT SIZE OF 0.5, THE SAMPLE 599 00:35:12,276 --> 00:35:14,779 SIZE IS 64 FOR EACH GROUP. IF 600 00:35:14,846 --> 00:35:20,017 WE WANT TO REDUCE THE SUBJECTS 601 00:35:20,084 --> 00:35:24,088 IN THE PLACEBO GROUP WE CAN USE 602 00:35:24,155 --> 00:35:31,262 THE RATIO OF 2:3 AND HERE WE GET 603 00:35:31,329 --> 00:35:39,704 1 84 AND -- 81. 604 00:35:39,770 --> 00:35:43,207 IT IS OBVIOUSLY HERE, THE4 AND 605 00:35:43,274 --> 00:35:43,975 IT IS OBVIOUSLY HERE, THE4 AND 606 00:35:44,041 --> 00:35:47,345 IT IS OBVIOUSLY HERE, THE54 AND 607 00:35:47,411 --> 00:35:49,380 IT IS OBVIOUSLY HERE, THE TOTAL 608 00:35:49,447 --> 00:35:51,349 NUMBER OF SUBJECTS THE TOTAL 609 00:35:51,415 --> 00:35:52,917 SAMPLE SIZE IS ALWAYS LESS THAN 610 00:35:52,984 --> 00:36:00,458 FOR THE UNBALANCED DESIGN. IT 611 00:36:00,525 --> 00:36:07,365 IS CRITICAL TO NOTE THAT THE 612 00:36:07,431 --> 00:36:11,369 TYPE ONE ERROR IS SET FOR SINGLE 613 00:36:11,435 --> 00:36:14,739 COMPARISON. YET WE DO MULTIPLE 614 00:36:14,805 --> 00:36:16,974 COMPARISONS. THE OVERALL TYPE 615 00:36:17,041 --> 00:36:26,083 ONE ERROR IS NOT 0.05 AIM. WITH 616 00:36:26,150 --> 00:36:26,951 INDEPENDENT COMPARISONS WITH THE 617 00:36:27,018 --> 00:36:30,087 TYPE ONE ERROR OF THE FAMILY 618 00:36:30,154 --> 00:36:32,823 ERROR RATE. CAN BE CALCULATE 619 00:36:32,890 --> 00:36:35,326 USING THIS FORMULA. FOR 620 00:36:35,393 --> 00:36:37,361 EXAMPLE, IF WE DO K -- IF WE DO 621 00:36:37,428 --> 00:36:42,433 5 INDEPENDENT COMPARISONS, AT 622 00:36:42,500 --> 00:36:46,203 SIGNIFICANT LEVEL OF 0.5 OF ALL 623 00:36:46,270 --> 00:36:54,011 TYPE ONE ERROR WILL BE 0.226 AND 624 00:36:54,078 --> 00:36:58,215 THE WEIGHT -- IF K GO TO 20 AT 625 00:36:58,282 --> 00:37:04,121 ALPHA EQUAL TO 0.05 THE TYPE ONE 626 00:37:04,188 --> 00:37:07,558 ERROR WILL BE 0.64 AND FOR 627 00:37:07,625 --> 00:37:11,862 THE -- IF WE DO 5 INDEPENDENT 628 00:37:11,929 --> 00:37:14,699 COMPARISON AND WE SAID ALPHA 629 00:37:14,765 --> 00:37:20,871 EQUAL TO 0.01 AND THEN THE 630 00:37:20,938 --> 00:37:22,907 FAMILY TYPE WE STILL HAVE 0.05 631 00:37:22,974 --> 00:37:25,643 FOR EXAMPLE THE STUDY INCLUDE 632 00:37:25,710 --> 00:37:27,979 ONE PRIMARY OUTCOME AND FIVE 633 00:37:28,045 --> 00:37:31,382 EXPLANATORY VARIABLES FOR THE 634 00:37:31,449 --> 00:37:36,954 ASSOCIATION TEST. TO KEEP FWER 635 00:37:37,021 --> 00:37:42,860 AT 0.05 WE SHOULD SET ALPHA AT 636 00:37:42,927 --> 00:37:46,364 0.01 IN POWER ANALYSIS. HOWEVER 637 00:37:46,430 --> 00:37:50,935 IT IS ESTIMATE 0 ON POWER OF 0.8 638 00:37:51,002 --> 00:37:54,405 AND 0.05 THE STUDY WOULD HAVE 639 00:37:54,472 --> 00:37:57,675 POWER OF 0.8 AFTER ADJUSTING THE 640 00:37:57,742 --> 00:38:01,345 P-VALUES FOR THE MULTIPLE TESTS. 641 00:38:01,412 --> 00:38:02,380 MULTIPLICITY CAN ARISE IN THE 642 00:38:02,446 --> 00:38:11,255 FOLLOWING SITUATIONS. IF THE 643 00:38:11,322 --> 00:38:12,657 PROTOCOL INCLUDES MULTIPLE 644 00:38:12,723 --> 00:38:19,163 PRIMARY HIYPOTHESES OR MULTIPLE 645 00:38:19,230 --> 00:38:20,998 PRIMARY OUTCOMES OR MULTIPLE 646 00:38:21,065 --> 00:38:23,534 EXPLANATORY VARIABLES OR 647 00:38:23,601 --> 00:38:25,536 SUBGROUP ANALYSES OR INTERIM 648 00:38:25,603 --> 00:38:28,906 ANALYSIS IF ANY OF THE ABOVE 649 00:38:28,973 --> 00:38:32,510 SITUATION WE NEED TO SET THE 650 00:38:32,576 --> 00:38:37,615 LEVEL TO 0.05 TO ACCOUNT FOR THE 651 00:38:37,682 --> 00:38:42,053 COMPARISON IN ORDER TO KEEP THE 652 00:38:42,119 --> 00:38:44,288 PLANNED POWER. OUR ANALYSIS IS 653 00:38:44,355 --> 00:38:50,327 BASED ON ASSUMPTIONS AND GUESSES 654 00:38:50,394 --> 00:38:52,263 AND MAYBE THE TWO UNDER -- 655 00:38:52,329 --> 00:38:57,068 OVERESTIMATE SAMPLE SIZE. SO IT 656 00:38:57,134 --> 00:39:00,638 IS RECOMMENDED THAT TO PLAN -- 657 00:39:00,705 --> 00:39:03,708 IF WE ARE NOT SURE -- ESPECIALLY 658 00:39:03,774 --> 00:39:07,344 WE ARE NOT SURE ABOUT THE 659 00:39:07,411 --> 00:39:09,213 COMPONENTS WE SHOULD -- WE NEED 660 00:39:09,280 --> 00:39:11,916 TO PLAN AN INTERIM ANALYSIS. TO 661 00:39:11,982 --> 00:39:16,220 ESTIMATE THE SAMPLE SIZE USING 662 00:39:16,287 --> 00:39:20,324 THE ESTIMATED DATA OF SIGMA. 663 00:39:20,391 --> 00:39:22,259 ALL TO RUN THE POWER ANALYSIS 664 00:39:22,326 --> 00:39:22,893 USING SEVERAL VALUES OF THE 665 00:39:22,960 --> 00:39:30,935 COMPONENTS. SUCH A AS THETA AN 666 00:39:31,001 --> 00:39:35,706 SIGMA AND TO GET MULTIPLE SAMPLE 667 00:39:35,773 --> 00:39:37,641 SIZE. THE SAMPLE SIZE BY 668 00:39:37,708 --> 00:39:38,642 CONSIDERING SOME OTHER FACTORS 669 00:39:38,709 --> 00:39:42,480 SUCH AS RESOURCES AND THE 670 00:39:42,546 --> 00:39:44,682 MINIMAL CLINICAL IMPORTANT 671 00:39:44,749 --> 00:39:47,118 EFFECT SIZE. SAY THERE'S A 672 00:39:47,184 --> 00:39:51,355 SAMPLE SIZE IS BASED ON 673 00:39:51,422 --> 00:39:53,824 ASSUMPTIONS AND GUESSES, WE 674 00:39:53,891 --> 00:40:04,335 CANNOT ESTIMATE PROCESSES 675 00:40:06,203 --> 00:40:10,207 PRECISELY. WE MAKE IT WITH THE 676 00:40:10,274 --> 00:40:13,210 OVER OR UNDERESTIMATED SAMPLE 677 00:40:13,277 --> 00:40:14,845 SIZE. HOWEVER, OVERESTIMATED 678 00:40:14,912 --> 00:40:16,347 SAMPLE SIZE IS TYPICALLY SAFE 679 00:40:16,413 --> 00:40:22,219 NEVER TERMS OF -- SAFE NEVER 680 00:40:22,286 --> 00:40:26,524 TERMS OF ACHIEVING RELIABLE 681 00:40:26,590 --> 00:40:27,358 SEARCH RESULTS. 682 00:40:27,424 --> 00:40:34,031 HOWEVER, THIS IS SEEN AS LESS 683 00:40:34,098 --> 00:40:34,598 RISKY THAN UNDERESTIMATING 684 00:40:34,665 --> 00:40:45,209 SAMPLE SIZE. NEXT I'M GOING TO 685 00:40:47,144 --> 00:40:47,845 PRESENT TWO EXAMPLES OF POWER 686 00:40:47,912 --> 00:40:54,285 ANNUAL SIS -- ANALYSIS. 687 00:40:54,351 --> 00:40:57,121 POWER ANALYSIS HAS TWO LOTS, ONE 688 00:40:57,188 --> 00:41:00,658 IS -- ONE IS SAMPLE SIZE 689 00:41:00,724 --> 00:41:01,292 ESTIMA 690 00:41:01,358 --> 00:41:03,360 ESTIMATION. THE OTHER ONE IS 691 00:41:03,427 --> 00:41:05,396 POWER ESTIMATION. THE FIRST 692 00:41:05,462 --> 00:41:08,833 EXAMPLE DRAWS UP SAMPLE SIZE 693 00:41:08,899 --> 00:41:10,000 ESTIMATION. THE SECOND EXAMPLE 694 00:41:10,067 --> 00:41:14,471 SHOW -- WILL SHOW THE POWER 695 00:41:14,538 --> 00:41:16,173 ESTIMATION. THIS EXAMPLE SHOW 696 00:41:16,240 --> 00:41:21,545 HOW TO USE -- THE PILOT STUDY OF 697 00:41:21,612 --> 00:41:25,082 THE SAMPLE SIZE. THE FIRST IS A 698 00:41:25,149 --> 00:41:27,084 PARALLEL GRIPE DESIGN. THE NULL 699 00:41:27,151 --> 00:41:29,286 IS BETWEEN THE TREATMENT OF 700 00:41:29,353 --> 00:41:31,856 ZERO. TO TEST THOSE HYPOTHESIS 701 00:41:31,922 --> 00:41:38,863 WE USE TWO SAMPLE T-TEST WHICH 702 00:41:38,929 --> 00:41:44,268 MEANS WE DON'T USE ANY 703 00:41:44,335 --> 00:41:46,237 COVARIANTS. THE CLINICAL SIZE 704 00:41:46,303 --> 00:41:49,940 IS 1.1 UNITE. THE PILOT STUDY 705 00:41:50,007 --> 00:41:53,944 IS CONDUCTED TO ASSESS 706 00:41:54,011 --> 00:41:56,981 FEASIBILITY AND ESTIMATE 707 00:41:57,047 --> 00:42:06,924 STANDARD DEVIATION. THIS IS 708 00:42:06,991 --> 00:42:10,027 ALSO THE PARALLEL GROUP DESIGN 709 00:42:10,094 --> 00:42:12,963 WITH SAMPLE SIZE 12 N GROUP. AT 710 00:42:13,030 --> 00:42:15,900 FIRST WE DID A TWO SAMPLE T-TEST 711 00:42:15,966 --> 00:42:19,236 USING THE PILOT STUDY. FROM THE 712 00:42:19,303 --> 00:42:24,775 TREATMENT GROUP, WE CALCULATED A 713 00:42:24,842 --> 00:42:29,947 SAMPLE AS THIS AND A SAMPLE 714 00:42:30,014 --> 00:42:32,750 STANDARD DEVIATION AT 2.6. WITH 715 00:42:32,816 --> 00:42:35,352 THE CALCULATED GROUP WE HAVE THE 716 00:42:35,419 --> 00:42:37,755 SAMPLE AS 4.2 AND THE STANDARD 717 00:42:37,821 --> 00:42:43,360 DEVIATION AT 2.15. THEY 718 00:42:43,427 --> 00:42:45,930 WEIGHTED EQUITY OF VARIANCE TEST 719 00:42:45,996 --> 00:42:55,072 AND GET P EQUAL 0.536 THIS 720 00:42:55,139 --> 00:42:56,373 INDICATES THAT THE VARIANCE FROM 721 00:42:56,440 --> 00:43:01,378 THE TWO GROUPS ARE SIMILAR. 722 00:43:01,445 --> 00:43:07,351 THEN WE CAN CALCULATE THE POOLED 723 00:43:07,418 --> 00:43:09,987 STANDARD DEVIATION. FOR THE 724 00:43:10,054 --> 00:43:11,889 TREATMENT EFFECT, WITH THE 725 00:43:11,956 --> 00:43:15,392 SAMPLE TREATMENT FACTOR IS 1.64. 726 00:43:15,459 --> 00:43:16,961 CALCULATED FROM THE DIFFERENCE 727 00:43:17,027 --> 00:43:19,063 FROM THE TREATMENT AND THE 728 00:43:19,129 --> 00:43:24,101 CONTROL GROUP. AND THE 95 729 00:43:24,168 --> 00:43:29,640 INTERVAL IS THIS. SINCE THE 95% 730 00:43:29,707 --> 00:43:30,541 INTERVAL INCLUDE ZERO THEREFORE 731 00:43:30,607 --> 00:43:33,944 THE DIFFERENCE IS NOT 732 00:43:34,011 --> 00:43:35,379 SIGNIFICANT FROM ZERO AT 733 00:43:35,446 --> 00:43:44,788 SIGNIFICANCE LABEL OF 0.05. SO 734 00:43:44,855 --> 00:43:47,958 THIS MEANS THAT WE ARE 95% SURE 735 00:43:48,025 --> 00:43:50,961 THAT THE TRUE VALUE OF DATA IS 736 00:43:51,028 --> 00:43:52,696 EXIST IN THE INTERVAL. TO 737 00:43:52,763 --> 00:43:56,200 CALCULATE THE SAMPLE SIZE WE 738 00:43:56,266 --> 00:44:06,810 NEED TO CALCULATE THE 80% UP AND 739 00:44:10,114 --> 00:44:13,117 GIVE IT OF THE DATA, OF THE 740 00:44:13,183 --> 00:44:16,220 SIGMA AND FIRST WE CALCULATE THE 741 00:44:16,286 --> 00:44:20,491 60% INTERVAL AS THIS AND THEN WE 742 00:44:20,557 --> 00:44:24,395 GET THE 80% UP ONE SIDED 743 00:44:24,461 --> 00:44:30,134 CONFIDENCE LIMIT AS THIS. WHICH 744 00:44:30,200 --> 00:44:32,970 MEANS WE ARE 80% SURE OF THE 745 00:44:33,037 --> 00:44:35,005 TRUE VALUE OF SIGMA. AND THEN 746 00:44:35,072 --> 00:44:38,509 WE DO POWER ANALYSIS USING THE 747 00:44:38,575 --> 00:44:43,347 TWO TAILED TWO SAMPLE T-TEST. 748 00:44:43,414 --> 00:44:50,087 AND POWER OF 0.8 AND ALPHA AT 749 00:44:50,154 --> 00:44:55,759 0.05 FOR THE SAMPLE SIDE 1.1 WE 750 00:44:55,826 --> 00:44:56,960 CALCULATE IT AS ONE HUNDRED 751 00:44:57,027 --> 00:44:58,228 GROUP. YET WE DO NOT KNOW THE 752 00:44:58,295 --> 00:45:03,367 EFFECT SIZE. WE CAN DO THE SAME 753 00:45:03,434 --> 00:45:07,204 THING AS WE ESTIMATE THE 754 00:45:07,271 --> 00:45:12,042 POPULATION STANDARD DEVIATION IS 755 00:45:12,109 --> 00:45:14,144 UP OF ONE SIDE INTEGRAL HERE WE 756 00:45:14,211 --> 00:45:18,582 USE LOWER ONE SIDED INTERVAL 757 00:45:18,649 --> 00:45:23,353 BECAUSE WE DON'T WANT TO 758 00:45:23,420 --> 00:45:25,289 UNDERESTIMATE THE OVER -- SORRY, 759 00:45:25,355 --> 00:45:29,059 WE DON'T WANT TO OVERESTIMATE 760 00:45:29,126 --> 00:45:35,365 THE EFFECT SIZE. AND THEN WE 761 00:45:35,432 --> 00:45:37,501 GET A 60% INTERVAL AND THEN WE 762 00:45:37,568 --> 00:45:48,112 GET 80% LOWER. THEREFORE WE ARE 763 00:45:51,315 --> 00:45:53,717 80% SURE THAT THE TRUE VALUE OF 764 00:45:53,784 --> 00:45:57,955 THE DATA IS GREAT THAN 0.81. 765 00:45:58,021 --> 00:45:59,857 BASED ON THIS WE GET THE SAMPLE 766 00:45:59,923 --> 00:46:06,663 SIZE AS 184. YOU SEE THE BIG 767 00:46:06,730 --> 00:46:07,564 DIFFERENCE BETWEEN THESE TWO 768 00:46:07,631 --> 00:46:15,339 SAMPLE SIZE. AND THIS PLOT WAS 769 00:46:15,405 --> 00:46:18,342 CREATED USING THE EFFECT SIZE 770 00:46:18,408 --> 00:46:22,146 FROM LOWER ONE SIDED LIMIT OF 771 00:46:22,212 --> 00:46:28,452 0.8 AND THE TWO ESTIMATE OF 164 772 00:46:28,519 --> 00:46:30,787 YOU WILL SEE THE SAMPLE SIZE 773 00:46:30,854 --> 00:46:34,224 DECREASE AS THE EFFECT SIZE 774 00:46:34,291 --> 00:46:42,266 INCREASES. THE SECOND EXAMPLE 775 00:46:42,332 --> 00:46:44,501 SHOW TO DETERMINE THE POWER WITH 776 00:46:44,568 --> 00:46:46,236 GIVING THE SAMPLE SIZE AND 777 00:46:46,303 --> 00:46:48,705 EFFECT SIZE. THE OBJECTIVE OF 778 00:46:48,772 --> 00:46:52,910 THIS STUDY IS TO ASSESS THE 779 00:46:52,976 --> 00:46:55,112 EFFECTIVENESS OF GROUP PSYCHO 780 00:46:55,179 --> 00:46:56,713 CHAIRPERSON FOR PATIENTS WITH 781 00:46:56,780 --> 00:46:59,950 CONVERSION DISORDER. THE 782 00:47:00,017 --> 00:47:05,422 DESIGN, FIRST WE WANT THE DESIGN 783 00:47:05,489 --> 00:47:06,490 AS PARALLEL GROUP DESIGN. SO 784 00:47:06,557 --> 00:47:09,293 AFTER THE INITIAL ASSESSMENT THE 785 00:47:09,359 --> 00:47:14,731 CD PATIENT WILL BE RANDOMIZED TO 786 00:47:14,798 --> 00:47:16,233 THE PSYCHOTHERAPY TREATMENT 787 00:47:16,300 --> 00:47:18,302 GROUP OR CONTROL GROUP. THE 788 00:47:18,368 --> 00:47:19,469 EFFICACY OF THE THERAPY WILL BE 789 00:47:19,536 --> 00:47:22,239 EVALUATED BY THE DIFFERENCE IN 790 00:47:22,306 --> 00:47:26,043 THE SCORE BETWEEN THE TREATED 791 00:47:26,109 --> 00:47:29,513 GROUP AND THE CONTROL GROUP. SO 792 00:47:29,580 --> 00:47:35,719 THE HYPOTHESIS IS 0. THE 793 00:47:35,786 --> 00:47:38,088 PRIMARY OUTCOME MEASURE IS THE 794 00:47:38,155 --> 00:47:41,158 SCORE OF ABILITY OF PARTICIPATE 795 00:47:41,225 --> 00:47:43,760 IN SOCIAL ROLES AND ACTIVITIES. 796 00:47:43,827 --> 00:47:45,395 THE HIGHER SCORES REPRESENT 797 00:47:45,462 --> 00:47:53,270 BETTER ABILITIES. BASED ON THE 798 00:47:53,337 --> 00:47:56,607 POPULATION AND THE STANDARD 799 00:47:56,673 --> 00:47:59,109 DEVIATIONS CAN HAVE THE MEAN OF 800 00:47:59,176 --> 00:48:01,445 50. THE INVESTIGATOR CONSIDERS 801 00:48:01,511 --> 00:48:05,882 A 15% CHANGE AS A CLINICAL 802 00:48:05,949 --> 00:48:09,753 IMPORTANT. THUS EFFECT SIZE IS 803 00:48:09,820 --> 00:48:15,492 7.5. THE INVESTIGATOR HAS 804 00:48:15,559 --> 00:48:17,561 RESOURCES ONLY 20 SUBJECTS SO 805 00:48:17,628 --> 00:48:18,629 THE POWER ANALYSIS IS TO 806 00:48:18,695 --> 00:48:21,365 DETERMINE THE POWER FOR GIVING 807 00:48:21,431 --> 00:48:23,367 SAMPLE SIZE OF 20 AND EFFECT 808 00:48:23,433 --> 00:48:28,739 SIZE OF 7.5. WE DID -- I DID 809 00:48:28,805 --> 00:48:32,542 THE TWO-SAMPLE T-TEST AT AN 810 00:48:32,609 --> 00:48:36,313 ALPHA 0.05 AND THE SIGMA 10. 811 00:48:36,380 --> 00:48:40,250 AND FOR THE SAMPLE SIZE 28. AND 812 00:48:40,317 --> 00:48:45,656 WE ESTIMATE THE POWER AS 0.355 813 00:48:45,722 --> 00:48:51,328 THIS MEANS FOR SAMPLE SIZE OF 10 814 00:48:51,395 --> 00:48:58,268 WE ONLY HAVE 0.355 THE 815 00:48:58,335 --> 00:49:02,939 DIFFERENCE OF 7.5 AT THE 816 00:49:03,006 --> 00:49:07,210 TREATMENT GROUP. FOR THE POWER 817 00:49:07,277 --> 00:49:12,282 0.8 THE SAMPLE SIZE IS 58 OR 29 818 00:49:12,349 --> 00:49:14,951 PER GROUP. THEN WE CHANGE THE 819 00:49:15,018 --> 00:49:17,854 DESIGN AS QUASI EXPERIMENTAL 820 00:49:17,921 --> 00:49:22,125 DESIGN. OF ONE GROUP PRETEST OR 821 00:49:22,192 --> 00:49:28,498 POSTTEST DESIGN. THIS DESIGN 822 00:49:28,565 --> 00:49:30,300 COMPARE WITH THE PRETEST DESIGN. 823 00:49:30,367 --> 00:49:32,803 SO UNDER THE INITIAL ASSESSMENT, 824 00:49:32,869 --> 00:49:36,340 ALL 20 PATIENTS WILL GET THE 6 825 00:49:36,406 --> 00:49:40,243 MONTHS OF PSYCHOTHERAPY. THEN 826 00:49:40,310 --> 00:49:44,047 WE WILL BE ASSESSED AGAIN 827 00:49:44,114 --> 00:49:46,249 POSTTEST. EACH SUBJECT SERVES 828 00:49:46,316 --> 00:49:51,321 AS HIS/HER OWN CONTROL. THE 829 00:49:51,388 --> 00:49:53,957 EFFICACY OF THE THERAPY IS 830 00:49:54,024 --> 00:49:56,293 EVALUATED BY THE DIFFERENCE IN 831 00:49:56,360 --> 00:49:57,627 THE SCORE BETWEEN THE PRETEST 832 00:49:57,694 --> 00:49:59,363 AND THE POSTTEST THE POWER 833 00:49:59,429 --> 00:50:02,265 ANALYSIS DETERMINE THE POWER FOR 834 00:50:02,332 --> 00:50:06,636 GIVING SOME AT 20 AND THE 835 00:50:06,703 --> 00:50:10,240 EFFECTS AS 7.5. BASED ON THE 836 00:50:10,307 --> 00:50:11,708 PAIRED T-TEST FOR THIS DESIGN 837 00:50:11,775 --> 00:50:14,945 AND ALPHA AT 0.05 STANDARD 838 00:50:15,011 --> 00:50:17,447 DEVIATION WE USE THE 10. HERE 839 00:50:17,514 --> 00:50:19,783 WE SEE THE CORRELATION IN THE 840 00:50:19,850 --> 00:50:21,852 SCORE BETWEEN PRE AND POSTTEST 841 00:50:21,918 --> 00:50:25,088 IS AT 0.5. AND FOR SAMPLE SIZE 842 00:50:25,155 --> 00:50:31,628 OF 20, WE GET THE POWER OF 843 00:50:31,695 --> 00:50:34,030 0.889. WHAT THIS MEANS THE 844 00:50:34,097 --> 00:50:36,366 STUDY WITH ONE GROUP DESIGN AND 845 00:50:36,433 --> 00:50:39,936 THE SAMPLE SIZE OF 20 WOULD HAVE 846 00:50:40,003 --> 00:50:45,542 POWER OF 0.89 TO DETECT THE 847 00:50:45,609 --> 00:50:50,680 SAMPLETIZE OF 7.5 AT A LABEL OF 848 00:50:50,747 --> 00:50:53,950 0.05. FOR POWER OF 0.8 THE 849 00:50:54,017 --> 00:50:56,019 SAMPLE SIZE IS 16. PLEASE 850 00:50:56,086 --> 00:50:59,089 NOTICE THE CONCLUSION FROM THIS 851 00:50:59,156 --> 00:51:01,992 STUDY IS SUBJECT TO BIAS. 852 00:51:02,058 --> 00:51:07,230 BECAUSE LACK OF RANDOMIZATION 853 00:51:07,297 --> 00:51:09,966 AND CONTROL GROUP. THEREFORE, 854 00:51:10,033 --> 00:51:12,836 THE CONCLUSION FROM THE QUASI 855 00:51:12,903 --> 00:51:15,138 EXPERIMENTAL -- THIS QUASI 856 00:51:15,205 --> 00:51:16,573 EXPERIMENTAL DESIGN IS MUCH, 857 00:51:16,640 --> 00:51:19,342 MUCH WEAKER THAN FROM THE 858 00:51:19,409 --> 00:51:27,851 PARALLEL DESIGN. AS WE COME TO 859 00:51:27,918 --> 00:51:32,789 THE END OF THE LEK CHUCTURE. ID 860 00:51:32,856 --> 00:51:34,257 LIKE TO EMPHASIZE THE IMPORTANCE 861 00:51:34,324 --> 00:51:37,894 OF POWER ANNUALYSIS NOT ONLY TO 862 00:51:37,961 --> 00:51:40,797 AVOID THE STUDY FROM BEING 863 00:51:40,864 --> 00:51:42,432 UNDERPOWERED OR OVERPOWERED BUT 864 00:51:42,499 --> 00:51:46,236 ALSO TO IMPROVE THE STUDY DESIGN 865 00:51:46,303 --> 00:51:50,640 SUBSTANTIALLY. BECAUSE 866 00:51:50,707 --> 00:51:51,341 PERFORMING POWER ANALYSIS 867 00:51:51,408 --> 00:51:53,610 REQUIRE INVESTIGATORS TO THINK 868 00:51:53,677 --> 00:51:56,813 THROUGH THE FOLLOWING ISSUES. 869 00:51:56,880 --> 00:51:59,349 HOW TO GENERATE THE HYPOTHESIS. 870 00:51:59,416 --> 00:52:03,520 WHAT ARE -- WHAT IS THE OUTCOMES 871 00:52:03,587 --> 00:52:05,222 WE SHOULD USE? WHICH ONE SHOULD 872 00:52:05,288 --> 00:52:08,525 BE PRIMARY? WHICH ONE SHOULD BE 873 00:52:08,592 --> 00:52:10,360 SECONDARY? AND WHAT TYPE OF 874 00:52:10,427 --> 00:52:15,131 VARIABLE SHOULD WE USE? 875 00:52:15,198 --> 00:52:18,134 CONTINUOUS, CATEGORICAL, AND 876 00:52:18,201 --> 00:52:22,239 ALSO THINK ABOUT DO WE NEED 877 00:52:22,305 --> 00:52:26,276 INTERIM ANALYSIS TO ESTIMATE THE 878 00:52:26,343 --> 00:52:31,748 COMPONENTS FOR POWER ANALYSIS? 879 00:52:31,815 --> 00:52:34,684 AND WHAT EXPERIMENTAL DESIGN 880 00:52:34,751 --> 00:52:38,822 SHOULD WE USE? OBSERVATIONAL 881 00:52:38,889 --> 00:52:41,725 STUDY? LONGITUDINAL, 882 00:52:41,791 --> 00:52:43,860 CROSS-SECTIONAL AND WHAT METHOD 883 00:52:43,927 --> 00:52:50,500 SHOULD WE USE? REGRESSION, AND 884 00:52:50,567 --> 00:52:55,338 DO WE INCLUDE THE COVARIATE OR 885 00:52:55,405 --> 00:52:57,774 NOT. AND WHAT RANDOMIZATION DO 886 00:52:57,841 --> 00:53:04,247 WE USE? DO WE USE THE COVARIATE 887 00:53:04,314 --> 00:53:06,316 RANDOMIZATION. AND FINALLY, FOR 888 00:53:06,383 --> 00:53:09,686 THE SIGNIFICANCE OF LEVEL. WHAT 889 00:53:09,753 --> 00:53:12,222 KIND OF SIGNIFICANCE SHOULD WE 890 00:53:12,289 --> 00:53:16,760 USE. IS BASED ON NUMBER OF 891 00:53:16,826 --> 00:53:19,362 COMPARISONS. THE NUMBER OF 892 00:53:19,429 --> 00:53:22,365 PRIMARY HYPOTHESES OR WE NEED -- 893 00:53:22,432 --> 00:53:29,472 OR WE JUST INCLIUDE SUBGROUP 894 00:53:29,539 --> 00:53:30,574 ANALYSIS AND FINALLY THE DROPOUT 895 00:53:30,640 --> 00:53:31,374 RATE DEFINITELY SHOULD BE 896 00:53:31,441 --> 00:53:40,183 CONSIDERED. AND IF YOU WANT TO 897 00:53:40,250 --> 00:53:42,819 GET TO THE POWER ANALYSIS HERE'S 898 00:53:42,886 --> 00:53:43,453 THE REFERENCES I USED FOR THIS 899 00:53:43,520 --> 00:53:47,591 LECTURE. AND THANK YOU VERY 900 00:53:47,657 --> 00:53:58,034 MUCH, ANY QUESTIONS? 901 00:54:02,872 --> 00:54:03,773 >> SLIDES WILL BE AVAILABLE AT 902 00:54:03,840 --> 00:54:04,240 SOME POINT AFTER THE 903 00:54:04,307 --> 00:54:14,551 PRESENTATION. 904 00:54:56,626 --> 00:54:57,360 >> I DON'T THINK I SEE ANY 905 00:54:57,427 --> 00:54:58,228 FURTHER QUESTIONS COMING 906 00:54:58,294 --> 00:55:03,366 THROUGH, YEAH. ALL RIGHT, THANK 907 00:55:03,433 --> 00:55:13,810 YOU, ALL, FOR JOINING. 908 00:55:13,877 --> 00:55:14,978 >> TIANXIA: THANK YOU VERY MUCH. 909 00:55:15,045 --> 00:55:15,578 ANY QUESTIONS, YOU CAN ALSO 910 00:55:15,645 --> 00:55:25,121 E-MAIL ME. 911 00:55:25,188 >> THANKS, TIANXIA.