WHO/HST/ESM/86.1 UPBATER AY 1997 PURLICAT LOR ASO w SAMPLE SIZE DETERMINATION Epidemiological and Statistical Methodology Unit World Heaith Organization Geneva, 1986 ee)WORLD HEALTH ORGANIZATION WHO /UST/ESM/86..1 ORGANISATION MONDIALE DE LA SANTE Saux ple a leedkee $¢ petted $ SAMPLE SYZE DETERMINATION A User's Manual Epidemiological and Statistical Methodology Unie World Health Organization Geneva Toe ksue of this document does not constitute formal publication. Ht should ot be reviewed, abstracted, quoted or translated without the agreement of the World Mealth Organization, Authors slone are respontible for views exprosted in Sioned stiles, Ce document ne constitue pas une publication. I ne doit five objet dovcun compte rendu ou résumé ni d'aucune citation ou raduetion tans Vautorisation de 'Orgenistion mondiale dels Santé. Las opinions exprimées dans. es. articles spés nfengogent que leurs auteursPREFACE INTRODUCTION TO THE TABLES Situation Sftuation Situation Situation Situation Situation Situation Situation Situation Situation Situation Situation Situation Sleuation Situation TABLES One Samphe Situations 1 2 3 Estimating population proportion with absolute precision Estimating population proportion with relative precision Hypothesis testing for a population proportion Two Sample Situations 4 5 Estimating the difference between two proportions Hypothesis testing for te population proportions Gase-control Study Situations 6 7 Estimating the odds ratio with stated precision llypothesis tests for the odds ratio Cohort: Study Situations 8 9 Confidence interval estimation of the relative risk Hypothesis vests for the relative risk Lot Quality Assurance Sampling Situations Acceptance of a population as having @ prevalence not exceeding a specified value Decision rule for "rejecting a sample" 10 un Incidence Race Study Situations 2 Bstination of incidence rates Hypothesis testing for incidence rete Hypothesis testing for two ineidonce rates Hypothesis testing for ewo inetdence rates in Follow-up studiesPreface A 1961 WHO document (¥810/2A/220.63) on sample size determination was reissued in revised form in 1973 with the title Adequacy of Sample Size {HS4/73.1), and has been Sn constant demand ever since. Recent changes in emphasis in the Orgenization’s activities have stimulated the production of a farther revision te include case-control studies and other surveys, This saw ‘edition fs intended to fill the need to have at one's elbow a document answering (Questions on adequacy of sample size, covering approaches most likely to be Saopted by health matagers in nonitoring and evaluating their programmes ‘Ths manual has been prepared by the unit of Epidemiological and Statistical Methodology in collaboration with the World Health Organization's Diatthowal Disease Control Programme, the Expanded Programe on Immunization and the Special Programme for Researeh and Training in Tropical Diseases. It Covers varying situations in which a health worker needs to select @ sample of a population under study. The document poses a mumber of typical questions which Realch workers ask statisticlans about the size of the sample they should Select. Je is hoped that the manual w£1l not only meet this need, but also provide insight into the methodology of solving probless about sample size. It Contains @ uininua mumber of tables expected to satisfy the denands of most health workers ‘The health worker planning a survey or study will ask "How large a sample do Lneed?". The answer will vary depending on the aims of the study. A Teazonable guess at the expected result is required, as well as how precise the result should be, the operational constraints, and other relevant information ‘the adequacy of the size of any sample depends on the scope of the study. For example, in a study of the ourative effect of a drug on a fatal disease such a3 the Acquired Inmune Deficiency Syndrome (ADS), where # single fowitive result would be important, sample size might be irrelevant, Dut in testing # new malaria vaccine, we are dealing with a disease which has & number Of known preventive measures of varying effectiveness, end the vaccine would have to be Eested of @ sufficiently large sample to permit comparison of its effect with that of other measures. A decision on sample sive aust also take into account the outcome category there are three possible outcome categories. The first is tho simple case yes/no, dead/alive, disabled/not dissbled, vaccinated/not vaccinated, existence we a health comittee/leck of a health committee. The second covers multiple, putually exclusive alternatives: religious beliefs, blood groups ete. For These tvo categories of outcone the data are generally expressed ss percentages or rates, The third category covers nunerical measurements: weight, height, age, blood pressure, body temperature ete. In this case the data are Summarized by means (averages) and variances or their derivatives. Regardless of population size, the absolute size of a minimup sample romaine almost the sane (except for very sual populations, i.e. less then 100). This inplies that if it is wished vo have a sample greater than the ainimun Fize, the sample should be ineveased by a multiple of the minimm (for exanple, from’ 150 to 200) rather than to a proportion of the whole population (for example, 5% co 108)sub In this manual 15 situations are considered, ‘They range from the minimun sample size for estimating the proportion of individuals in a population possessing a specified characteristic, to the minimum sample cize for testing the difference between incidence rates. For each situation at least one typical example hes been vorked out, Random sampling is assumed fox all. examples, so that if the sample {s not selected in a statistically random manner, the tables carmot be used. ‘The information needed to determine the minimum sample size is specified for each situarion, Thi information is identified in each example and then used to read the solution from the tables. Except for Situation 13, no computations are given and no theoretical statistical background 4s required, Brief notes are given to clarify sone of the points, The marusl is designed co be used in "cook-book" fashion. The user who wishes to learn the basis of sample size determination is referred to any standard textbook on statistics, or to the proposed revision of the WHO publication: "Adequacy of Semple Size”WHO/HST/ESM/86.1 Page 2 1. OWE SAMPLE SITUATIONS L.a. Sample Size for Estimating the Population Proportion ‘Two situations axe dealt with in chis subsection: estimation of a Proportion with absolute and relative precisions SITUATION 1: Estimating population proportion with absolute precision (Table 1 and Notes on page 3) Minimum sample size for estinating, with absolute precision, the proportion of individuals in @ population possessing a characteristic. Roquired Information and Notation (1) anttetpated population proportion (P]. (Soe Note (i)) (41) Confidence level (100(1-a)4] (441) Absolute precision (d] (1.e., in percentage points) required on either side of the proportion. Example 2 A local health department wishes to ostinate the prevalence rate of Aubeteulosis infection among children under five years of age in its locality Bow many children should be included in the sample so that the yate may be eatimated to within 5 percentage points of the true value with 959 confidence if ft 4s known that the true rate is unlikely to exceed 20%? Solution: (4) Anticipated population proportion (per cent): 208 (44) Confidence Level: 958 (441) Absolute precision (from 15% to 254) 5 percentage points From Table la in the colum headed "0.20" and the row headed "5" te 1s found that a sample size of 266 would be needed. (See Note (ii)} Example 2: The national progranse of immmization seeks to estimate the proportion of children recetving appropriate childhood vaccinations in the country. Assuming @ simple random sampling strategy, how many children must be studied if tne resulting estimate is to fall within 10 percentage points of the true proportion with 958 confidence? (It {2 not possible to make any assumption regarding che vaccination coverage.) Solutios (2) Antietpaced population proportion is unknown in this case, assume: 508 (See Nore (iv) (41) Confidence Level 958 (iii) Absolute precision (from 408 to 608): 10 percentage pointe From Table La in the column headed "0.50" and the row headed *10" a canple size of 96 would be required, (See Note (ii4))WHO/HST/ESH/26.1 Page 3 ne ety gh a {See Note (v)) A rough estinate will usually suffice, For a situation in which no such anticipation is possible a figure of 0.5 should be used as in Exemple 2, If the antieipated proportion fe given as a range, one should use the value closest to 0.5 If studying these many children is unrealistic wich respect to time and woney, the investigators should lover thair requirenents of confidence to, perhaps, 908. In this case, from Table 1b, a sample size of 174 (in the column headed "0.20" and row headed "5*) would be necessary. Simple random sampling is unlikely co be the sampling method of choice ia an actual fleld survey. As a result a larger sample size ie likely to be needed due to the “design effect". For example, 1f 2 cluster saupling strategy was to be used, the design effect might be estimated as 2. This would mean that in order to cbtain the sane precigion twiee as many individuals must be studied with cluster sampling as with the simple random sampling strategy. ‘The sample size vequixed will be largest when P is equal to 0.5 as can be seen in Tables 1a and tb, Thus the value 0.5 may be used for P in order to be on the safe side "z1.q" represents che nunber of standard errors from the mean and is a function of the confidence level. This term will be fully explained in the propsed revision of the VHO publication: “adequacy of Sample Size".wo ust /esm/e6,2 Page 4 SITUATION 2: Estimating population proportion with relative precision (Table 2 and Notes on page 5) Min{min gomple size for estimating, vith relative precision, the proportion of individuals in « population possessing a characteristic Required Information and Notation (2) Antietpated population proportion [P] [See Note (4)] (44) Confidence level [100(1-a)8) (If) Relative precision [€] Exanple 3: ine national programme of inminization sacks to estimate the proportion of shuzdren receiving appropriate childhood vaccinations in the country, Assuniog @ Simple random sempling strategy, how many children must be studied 4f the resulting estimate is to fall within 10¥ (not 10 percentage points) of the true proportion with 958 confidence? (The vaccination coverage ia not expected to he below 508.) Solues (4) Anticipated popularion proportion is unknown therefore, assune 508 Gt) Confidence level 958 (it) Relative precision (from 45¢ to 558): 10% (of 508) From Table 2a in the colunn headed "0,50" and the row headed "10" it ig found that @ sample size of 385 would be needed. [See Notes (11) and (411)}) Example 4 How large @ sample vould be required to estimate the proportion of pragnant women in the population who seek prenatal care within the first trimestes of Pregnancy to within 58 of the true value with 958 confidence? It is estinated thac the percentage proportion of women seeking such care will be somechers between 25% and 408 ution: (4) Anticipated populacion proportion (per cent): 258 to 40% it) Congidance level 958 Relative precision: 58 (of 258 to 408) Presents the following sample sizes, in the row headed "5", for the dlfferent: population proportions in the range 25% - 408; Proportion Sample size Poe aI aelOd: 3586 2854 2305, Therefore a study might be planned with roughly 4610 wonen to satisfy the study objectives. If, however, this number is too large then a smaller sexple afce RAght be used with @ loss of either precision or confidence or bothWHO/HST/ESM/86.1 Page 5 Sample size formula for Situation nm 22) .g(1-P)/ (272) (2) {See Notes (iv) & (v)) ‘The expected proportion used for the sample size computation should be as conservative as possible. If studying these many children is unrealistic with respect to time and money, the investigators should lover their requirements ‘of confidence to, perhaps, 908. In this case, from Table 2b, 2 mple size of 271 (in the column headed "0.50" and Tow headed "10") would be necessary. Simple random sampling ie unlikely to be the sampling mothod of chofce in an actual field curvey. Ac a result a larger semple Size is likely to be needed due to the “design effect". For exanple, if a cluster sapling strategy was to be used, the design effect might be estimated as 2. This would mean that in order to obtain the sane precision twice as many individuals must be studied vith cluster sanpling as with the simple random sampling strategy. Formula (2) te equivalent to Formula (1) with d replaced by (eR). szyq" represents the nusber of standard errors from the mean and Is a function of the confidence level. This term will be fully explained in the propsed revision of the WHO publication: "adequacy of Sample Size"WHO/HST/EsM/86.1 Page 6 1.b. Sample Size for Hypothesis Testing for a Population Proportion STTWATION 3: Hypothesis testing for « population proportion (Tables 3, 4 and Notes of page 8) Minimum sample size for testing the hypothesis that the proportion of Andivicuals in s population possessing a characteristic is equal to a particular value, Required Information and Notation (2) Hypothesized value of the population proportion [Po] (44) True value of the population proportion [Pq] (it) Level of significance [100(a)%] (iv) Power of the test [100(1-A)8] (w) Alternative hypothesis: either Pq > Py or Pq < Py (for one-sided test) or Pam (Zor tworsided test) Example 5: The five-year cure rate for a particular cancer is reported in the Literature co be 508. what minimum sample size would be needed to have a 908 probability of showing that a cure rate of 40t is not significantly different from the rate of $08 at the 58 level of significance? Solution: (i) Hypothesized cure rate: 408 (it) True cure rate: 508 (141) Level of significance: 58 (tv) Power of the cast 908 () Alternative hypothesis: cure rate is grester than 408 From Table 3 (for the level of e{gnificance of St, and a power of 90%) [See Note (1)] the sample size is given in the column headed "0.40" and row Readed "0.50" as 210. [See Note (11)] Exanp! Previous surveys have denonstrated that the usual rate of dental caries among school children in a particular conminity is about 258. How many childven should be included in a new survey if it is desired to be 90% sure of detecting 2 dental caries rate of 20¢ at the 5% level of significance? Solution (1) Hypothesized caries rate: 258 Gi) True caries race 208 (iL) Level of significance: 38 (iy) Power of the test 908 (v) Alternative hypothesis: caries rate is less t From Table 3 (for the level of significance of 5%, and a power of 908) the sample size is given in the column headed "0.25" and row headed "0.20" as 601 (See Note (414)]WHO/HST/ESM/86.1 Page 7 Example 7 Suppose the success rate for a surgical treatment of @ particular heart condition 1s widely reported in the literature to be 70%. A now medical treatment has been proposed which is alleged to offer equivalent treatment success. A hospital without the necessary surgical facilities or staff has decided to use the new medical treatment on all new patients presenting with this condition. How many patients mst be studied to test the hypothesis that the success rate of the new Creatment method is 708 against an alternative hypothesis that it is not 708 at the 5¢ level of significance? It is desired to have a 90 pover of detecting a difference in the proportion of success of 10 percentage points or greater? Solution: (4) Hypothesized success rate: 708 (44) True success rate: B08 or 608 (it) —Lavol of significance: 58 (iv) Power of the test: 908 (v) Alternative hypothesis: suecess rate is not equal co 70 Frou Table 4 [See Notes (1) and (iv)] for the level of significance of 54, and 4 power of 50 with absolute difference of 10 percentage points, the sample size is given in the column headed "0,30" and row headed *0,10* as 233. [see nore (v)] Example & ‘The proportion of patients seeking prenatal care in the first trimestar of pregnancy is estimated to be 40% by figures released by the provincial departaent of health. Health officials in another province are interested in comparing their success at providing prenatal care with the published figures. How many woxen should be sampled in order to teat the hypothesis that the coverage rate 1s 40$ against the alternative chat it is not 408? It is desired to be 908 confident of detecting a difference of as much as 5 percentage pointe at the 5% level of significance. Solution’ (1) Hypothesized success rate: 408 (1) True success rate 358 oF 458 CHD) Level of significance 58 (iw) Power of the test 908 () Alternative hypothesis: coverage rate is pot equal te 40% From Zsble 4 [See Notes (1) and (iv)] for the level of significance of St, and a power of 508 with absolute difference of 5 percentage points, the sample size is given in the column headed "0.40" and row headed "0.05" as 1022WHO/HST/ESH/86.1 Page 8 Sample size formlae for Situation 3: For L-elded test > (21 .a/[Po(t-Po)} + 21-g/(Pa(-Pa))12 / (Po-Pg)? For 2-sided test: Bo {21 -a/2/[Poll-Po)] + zy.p/[Pa(t-Pa)]}2 / (fo+P a)? (a) [See Note (vL)) Tables giving other choices for level of significance and power may be found in the proposed revision of the WHO publication “Adequacy of Sample Size" This is a one-sided cost since we are interested in rejecting the null hypothesis only when the true cure rate {s larger than the estimated one, If we use this sample size, and 1 the actual caries rate is less than 208, then the power of the test will be larger than 908, 4.6. we are more than 904 likely to detect that rate. Table 4 is used in this example since this is a two-sided test, That is, we are interested in rejecting the null hypothesis if the true success rate is different from the estimated success rate in either direction. Columns of Table 4 correspond to the hypothesized proportion whenever this proportion is less than or equal to 508. tthen the Proportion is greater than 50% the table is used by computing the complement (1-P) and using it as the colunn value. TFi-a/2") “Zl q" and "21.9" represent the nunber of standard exrors from the mean and are functions of the confidence level and pover respectively. "a" represents the type-I error and @ Toprosents the type-II error. These terms will be fully explained {n the above mentioned publication.wHlo/HST/ESH/86.1 Page 9 ‘TWO SAMPLE STTUATIONS 2.a, Sample Size for Estimating the Difference between Two Proportions SETUATION 4: Estinating the difference between two proportions (Table 5 and Notes on page 10) Minimun sample size for estimating the difference between two population proportions with absolute precision. Requived Information and Notation (2) Anticipated population proportions (P] and Po] GS) Configence Level [100(1-a)8] GAL) Absolute precision [4] (i.e., in percentage points) required on either side of the true value of the differences, (iv) An intermediate calculation [V ~ P,(1-P1)4P2(1-P2)] which may be Yooked up directly in Table 5a. Example 9 What size sample should be selected in each of two groups in order to estimate a risk difference to within 5 percentage points of the true difference with 95% confidence, when no reasonable estinate of Pj and P2 can be made? Note ()] Anticipated populetion proportions (per cont): 50%, 50% Confidence level: 958 Absolute precision: 5 percentage points Interaediate value: [fron Table 5a, See Note (11)] 0.50 From Table Sb in the column headed "5" and the row headed "0.50" iv is that in a pilot study of 50 agricultural workers in an irrigation project, {t was observed that 40% had active schistosomiasis. A similar pilot study of 50 agricultural workers outside the project demonstrated that 32% had active schistosomiasis Infection, If we would like to carry out a larger study to estimate the true schistosomiasis risk difference to within 5 percentage points of the true value with 958 confidenee, how many people must be studied in each of the two groups? Solution: (4) Anticipated population proportions (per cent): 408, 328 (4) Confidence Level: 958 (Gii) Absolute precision: 5 percentage points Civ) Intermediate value: [from Table Sa, See Note (ii)] 46 From Table 5b in the column headed "5" and the row headad "0.46" it is found that e sample size of 707 would be needad in each group.WHO/HST/ESM/86.1 Page 10 Sample size formuls for Situation 4: nm 22, gy2iPy(1-Py) + Po(1-Pp)1/a? «sy [See Note (144) For any absolute precision (d), the sample size selected will be largest when both Py and Pp are equal ro 50%, ‘The intermediate calculation Py(1-P1)+P7(1-P2) as used in equation (5) may be looked up directly in Table Sa, by entering the column corresponding to Pz (or {ts complement) and the row corresponding to Py (or its complement). "21 9/2" tepresente the number of standaxd erzors from the mean and is « function of the confidence level, This term will be fully explained in the propsed revision of the WHO publication: "adequacy of Sample Size",WHO/HST/BSM/86.1 Page 11 2.b. Sample Size for Hypothecic Testing tor Two Population Proportions srmvation 5: Hypothesis testing for two population proportions (Tables 6, 7 and 8, and Notes on pages 13) Minimum semple size for testing the hypothesis that tvo population proportions are equal. {For rare events, see Note (i)] Required Information and Notation ()—Hypothesized value of the difference between the population proportions [P1-P2 -0] (41) Actual values of the population proportions (Py and Pz] (iii) Level of significance [100(a)%} Gv) Power of cha test [100(1-A)] Alternative hypothesis: either Py-P2>0 or P1-P2<0 (for one-sided test) Py-Pp 0 (for two-sided test) Example 11: An investigator believes that the proportion of patients who will develop complications with one type of surgery is 5 while the proportion of patients developing complications with a second type of surgery will be 15%. How large should che sample size be in each of the two groups of patients to detect whether the second procedure has a complication rate which is significantly higher than the first at the 58 level of significance with a power of 9087 Solucto: (i) Hypothesized difference tn complication rates on (Gi) -Aetuall complication rater: 58, 154 (41) Level of significance: Se (iv) Power of the test: 908 (y) Alternative hypothesis: riek difference (first minus second) ie less than 0% From Table 6 (for the level of significance of 5%, and a power of 908) (See Note Gi}] the sample size ic given in the column headed "0.05" and row headed "0,15" as 153 in each group. [See Noce (iii)] Example 12 In = pilot survey in a developing country, an opidentologist compared & sample of 50 adult subjects suffering from a cextain neurologic disease to a sample of 50 conparable control subjects who were free of the disease. Thirty of the subjects with the disease (608) end 25 of the controls (500) vere involved in fishing, or fishing-related occupations. Assuming that the proportion of people involved in fiehing, or fishing-related occupations in the entire population i¢ similar to that observed in the pilot survey, how many subjects should be Included in a larger study in each of the tuo groups to have 90% confidence of detecting the true difference between the groups if we will test the hypothesis at the 58 Level?WO /HST/89M/86.1 Page 12 Solution: (4) Hypothesized difference in rates of exposure to fishing 08 (43) Actual exposure rates 608, 508 (iis) Level of significance: 58 (Av) Power of the test: 908 () Alternative hypothesis: risk aifference ts not equal to 08 From Table 7 (for the level of significance of 5%, and a power of 908) {See Note (iv)] ‘the sample size is given in the colunn headed "0.40" and tow headed "0.10" as 519 {n each group. [See Note (v)] Example 13 Tvo communities, differing in exposure to a particular environmental chemical pollutant, ate to be identified for screening for a certain type of cancer which is believed to be associated with this particular pollutant. A screening progranne for early detection of the cancer will be used on ald Permanent resident adults over 35 years of age in the selected commmities ‘The general incidence rate of the cancer is believed to be about 20 per 100 000 population (9.0002) in this age group. If the incidence in the exposed community 18 about 50 per 100 000 population (0.0003) how many people should be followed up in each community to have a 90% probability of detecting « significant difforence in the incidence rates at the 5¥ level of significance? Solution: (1) Hypothesized difference in cancer rat oe Gi) Actual cancer rates: 05%, .028 (4if) Level. of significance 38 (iv) Power of the test: 908 (v) Alternative hypothesis: risk difference (first minus second) is greater chan 08 From Table 8 (for the level of etgnificance of St, and a power of 90%) [See Note (vi)] the sample size is given in the colums headed "0.0002" and row headed "0.0005" as 63377 in each group. [See Note (1ii)] ample size formulae for Situation For one-sided test: mm (2) qdf2B(1-P)] + 2) g/EPL(L-Py) + Po(1-P2)])2/(Py-PQ)2 (6) For two-sided test: mm (ayeqs2V{2P(-P)] + 2y-g/TRL(L+Py) + Po(1-P2))12/(Py-Pg)? (7) For one-sided test with rare events: no GLa + 21-9)2/[2(arestn/P, - arcein/Py)2] @) [See Notes (vii), (vill) and (ix)]‘WHO/HST/ESH/86,1 Page 13 For rare events, refer to éxample 13. ‘This is a one-sided test since ve axe interested in rejecting the null hypothesis only when the actual éifference (first minus second) in complication rates is significantly less chan zero. This table is accessed by looking up the smaller proportion in the colunn and the larger proportion in the row. Table 7 is used in this example since this ie @ two-sided test That is, we are interested in rejecting the mull hypothesis if the actual difference in the rates 4 not equal to zero i.e, the First rate is greater than the second, or vice versa ‘The necessary sample siz0 is found at the intersection of the column corresponding to the smaller of Pj and (1-P)) and the row corresponding to Py-P2. Whenever the event under consideration is so rare that the senple Size produced using Table 6 results in the product nPy or nPz being less than 5, it is recomended that Table 8 be used instead. "21.9/2". "21a" and "21g" represent the number of standerd errors from the mean and are functions of the confidence level and power respectively. "a" represents the type-I error and f represents the type-I error, These terms will be fully explained in the proposed resivion of the WHO publication: “Adequacy of Sample Size". The quantity P ic computed as the average of Py and Pp. When computing n using Formula (8), Py and Pp should first be converted to radians.WHO/AST/ESH/86,1 Page 14 CASE-CONTROL STUDY SITUATIONS 3.2, Sample Size for Estimating the Odds Ratio with Stated Precision The odé0 ratio is the ratio of the probability of occurrence of an event to that of nonoccurrence. For example using the following notation: Exposed Unexposea Disease a » Wo disease © a the odds ratio is ad/oc. [See Note (i)] SITUATION 6: Estimating the odds ratio with stated precision (Table 9 and Wotes on page 15) Minima canple size for estimating the population odds ratio with relative precision. Required Information and Notation (2) Two of the following quantities should be known. [See Notes (41) and (144)] Anticipated probability of “exposure for persons with the disease” [Pj], Anticipated probability of “exposure for persons without the Sisease" [P2*] or Anticipated odds racio (OR) (44) Confidence Level. [100(1-2)] (4L1) Relative precision {¢] required, Example 1 About 308 of a population in a defined area are believed to be using water from suspected contaminated xources. A case control study of cholera 1s to be undertaken in the area to estimate the population disease odds ratio to within 258 of the true value which is believed to be approximately 2. What sample sizes would be needed in the case and control groupe? Soluttor (8) Anticipated probability of “exposure given disease” [See Note (4i)] 2 Anticipated probability of “exposure given no disease" 30% Anticipated odds ratio 2 (41) Confidence level [100(1-0)8] 58 (i4i) Relative precicion [4] required. 25 From Table 9¢ in che column headed "2" and the row headed "0.30" te is found that a sample size of 408 would be needed in each group [See Note (iv)]WHO/HST/BSM/86 1 Page 15, Sample size formula for Situation 6: nw 22a cqy2tB/UPE*CL-Pr*)] + 2/ [Ro (LeBg")11/[10C2-€) 1? ” Notes: wo [See Note (v)) When the number of people affected by the disease is small relative to the number of people unaffected : em (ate) and 4 = (bid) approximately, thus the probability of "exposure given no dise approxinated by the “overall exposure rate". In solving these problens there are three parameters (Py*, Fo* and OR) but the only two parameters necessary for accessing Tables 9a-9d are Py” and OR. Zf P]* and Py” axe given, chen OR may be determined by: oR = (ey*7Ch-By*D1/(24/A-P2"D) Similerly, if Py* ané OR are given, thon Pg ~ (Py) /1OR(L-Pa")4P1"*] “Exposure” refers to any variable suspected of being related to outeome, "Disease" refere to any outcome. Tables 9a-94 present 4 levels of precision for the sane level of Zonfidence (95%). More extensive tables will be available in the proposed revision of the WHO publication: "Adequacy of Sample Size" "224.q/2" represents the number of standard errors from the mean and {2 4 function of the confidence level. This term vill be fully explained in the above mentioned publication. The natural log is denoted "In"lO /#ST/B5M/86,,1 Page 16. 3-b, Sample Size for Hypothesis Tests for the Odds Ratio SITUATION 7: Hypothesis tests for the odds ratio (Table 10 ana Notes on page 17) Minimum sample size for testing the hypothesis that the population odas ratio is equal to one Required Information and Notation () Two of the following quancities should be known [See Notes (1) and (11)] Anticipated probability of “exposure for pereons with the disease” [)"], Anticipated probability of "exposure for persons without the disease” [P7*) or anticipated odds ratio [OR] Level of significance (100(a)4) Power of the test [100(1-)3} Example 15: In a given large community it is believed that about 308 of the people use water from contaminated sources. If @ case control study is to be usterteken Cp Sse, the hypothesis that the odds ratic for cholora in the coumnity is 2, at the 58 level of significance and a power of 908, how many cholera casen and controls should be studied? Solution: (4) Anticipated probability of “exposure given disease" (See Nore (1)} ? Anticipated probability of "exposure given no disease" [See Note (111)} 308 Anticipated odés ratio 2 (i) Level of significance 58 (41) Power of the test 908 Fron Table 10 in the column headed "2" ané the vow headed "0.30" dt ts Found that a sample size of 176 would be needed in each group [See Note (iv) jample formula for Situation 7: Mm ete ad[2P2* (1-22) }+24 pd 1Pp*(L-Py*)+Po*(1 PQ") | )2/(Py*-Py4)2 ao) [See Note (v)}WHO /HST/ESH/86.1 Page 17 In solving these problens there are three parameters (P)*, Po* ‘and OR) but the only two parameters necessary for accessing Table 10 are Py” end OR. If Py” and P* axe given, then OR may be determined by on = (PyX/-PY*D1/1PA/A-P") 1+ Similarly, L€ PL* and OR are given, then py* = (ey) /fora-By +P A"). “Exposure” refers to any variable suspected of baing related to outcome, “Disease” refers to any outcone the “exposure rate given no disease" is here approximated by the roverall exposure rate" teble_10 presents sample sizo for a ~ 5% and pover ~ 90% gutensive tables will be available in the proposed revision of the WHO publication: "Adequacy of Sample Size". “z}.q/27 and "21." represont the muber of standard exrors from the mean and are functions of the confidence level and power respectively. "a" and "A" refer to the type-I and type-II errors respectively, these terms will be fully explained in the above mentioned publication.WHO /ST/ESH/86.1 Page 18 4. COWORT STUDY SITUATIONS ‘4a. Sample Size for Confidence Interval Estimation of the Relative Risk Relative risk is generally éefined as the ratio of the risk of discase or Seath among the exposed group of people to the unexposed; where risk is the chance that an event will occur. SITUATION 8: Confidence interval estimation of the relative xisk (Table 11 and Wotes on page 19) SSS Notes on page 19) Mininum sample size for estimating the relative risk of a disease in a population with reletive precision. Required Information and Notation (2) Tuo of the following quantities should be known: [See Notes (1) and (1i)] Anticipated probability of “disease for persons with exposure" (Pi). Ancletpated probability of "disease for persons with no exposure” Pp] or Anticipated relative risk (RR) (34) Gonfidence Level [100(1-a)#} (it) Relative precision (¢] required Example 16: Suppose it is known that a disease is present in 208 of the unexposed group. What sample size would be needed in each of the tvo groups, exposed and Bon-exposed, to estimate the relative risk to within 50% of the true value, which {6 believed to be approximately 2, with 958 confidence. Solution G) Anticipated probability of "disease given exposure [See Note (¢)) Anticipated probability of "disease given no exposure Anticipated relative risk (ii) Confidence evel (100(1-a)%) 5a (Ut) Relative precision [€] required 508 From Zable 11d in the colum headed "2" and tha row headed "0.20" it ts found that a sample size of 44 would be needed in each group. [Seo Note (iii)] Sample size formila for Situation fm 2) 9/2! (1-PL)/PyHCL-Pp)/PpI/TAm(2-6) }2 ay [see Note (1v))wHo/HsT/ESH/86.1 Page 19 In solving these problexs there are three parameters (Pi, P2 and RR) but the only two necessary Zor accessing Table 11 are Pg and RR, If Py and Pp are given, then RR may be determined by RR = PL /?2 similarly, if PF] and RR are given, then Pp = P/aR “Exposure" refers to any variable suspected of being related to outcome, "Disease" refers to any outcome Liaslld present 4 levels of precision for the sane level ‘af confidence (95%). More extensive tables will be available in the proposed revision of the WHO publication: "Adequacy of Sample Size". "2219/2" represents the mumber of standard errors from the mean end is a function of the confidence level, This term will be fully explained in the above mentioned publication. The natural log is denoted by "In".WHO /#S1/B9M/86,.1 Page 20) 4.b. Sample Size for Hypothesis Tests for the Relative Risk SITUATION 9: Hypothesis teste for the xelative risk (Table 12 and Notes on page 21) Minimum sample size for testing the hypothesis that the population relative risk is equal to one Required Information and Notation (1) Two of the following quantities should be known: [See Notes (4) and (1i)} Anticipated probability of "disease for persons with exposure" (7), Anticipated probability of "disease for persons with no exposure” [Pa] or Anticipated relative risk [RR] (ii) Level of eignsfieance [100(a)¢) Gi) Power of the cest [100(1-A)8} Example 17 Two competing therapies for a particular cancer are to be evaluated by a cohort study strategy in a multicentre clinical trial. Patients are randomly allocated to either treatment A or B and are followed-up for 5 years to record any recurrence of the discase. Treatment A 4s a new therapy which will be widely adopted if it can be demonstrated that it cuts the risk of recurrence after 5 years in half (1.e., RR =.5) over what is currently observed with Treatment B (35% recurrence is presently common). How many patients should be studiod In each of the two treatment groups to be 908 confident of rejecting the hypothesis correctly if the test is to be performed at the 5¢ level of significance olution () Anticipated probability of "recurrence given treatment A" [See Nove (1)] 2 Anticipated probability of "recurrence given treatment BY 35% Anticipated relative risk 0.5 (45) Level of significance 58 Git) Power of the cese 908 From Table 12 in the column headed "0.5" and the row headed "0.35" ic te found that a semple size of 131 would be needed in each group. [See Nore (i11)] Sample size forma for Situation 9: Sample size formula for Situation 9: Bo (21 -a/2/[2PA-P) 42z-p/1Py (L-Pyy*Pg(1-P9))12/(Py-P2)2 (2) [See Note (iv)WHO /asT/ESM/86.1 Page 21 a solving these problems there are three pareneters (P1, Pz and HR) but the only two necessary for accessing Table 12 ate Pp and BR. If P1 and Po are given, then RR may be determined by RR - Py/Pa. Similarly, if Py and BR ore given, then Pp ~ PL /RR Ie should also be noted that RR has a renge of 0 to 1/P2. Nexposuve" refers to ony variable suspected of being related to outcone. "Disease" refers to any outcome. ‘table 12 presents sample size for @ ~ 58 and pover = 908. More satensive tables will be available in the proposed revision of the WHO publication: “Adequacy of Sample Size”. sz1.q/2" and "zy." represent the number of standard errors frin’tke mean end are functions of the confidence level and power respectively. "a" and "A" refer to the type-I and type-1T Etrors respectively. These terms will be more fully explained in the above mentioned publication. 2 denotes the average of Py and Pp.WHo/tsT/ESM/86.2 Page 22 ‘LOT QUALITY ASSURANCE SAMPLING creel tion 20; qfteeptance of a population as having a prevalence not exceeding specified valu (Table 13 and Notes on page 23) Devornine the minimum sample size which should be selected from a population of a given size so that if no more than a specified nunbex of elements vith a particular characteristic are found in the sample, the Population can be accepted as having 4 prevalence of not more than @ specified value. [See Note (i)) Required Information (2) Antictpaced population prevalence P (44) Population size 8 (114) Acceptable characteristic threshold in the sample a (iv) Confidence level 100(1-a)a Example 18: In @ school community of 2500 children how many children should be exantned for aalaria parasitaemta So that if no more chan two are found to have melerie Parasites it could be concluded, with a probability of 95%, that the malaria Prevalence in the school is no more than 108? Solution (2) Antictpated population prevalence Loe (44) Population aize 2500 (441) Acceptable malaria cases threshold in the sample 2 (x) Gonfidance level 958 Fyom Table 13¢ in the colusn headed "0.10" and row headed 2500 it fs found chat 61 children would be sufficient to exanine. If no more than tws of cree are found to be positive then it could be concluded that the prevalence af walaria in the school is no more than 10¥ Sample size formilse for Situation 1 Solution for n in the inequalities: an where N= NP, for Einite population; or a Tap Pepyex x0 for infinite population.WHO/HST/ESM/86.1 Page 23 SITUATION 11: Decision rule for "rejecting a sample" (Table 14 and Notes on page 23) Determine the combination of sample size and threshold value of the characteristic in the sanple for testing the hypothesis that the population hes a particular proportion of individuals with the characteristic. (The threshold Value is the gaxinun acceptable nunber of individuals with the characteristic for acceptance of the hypothesis.) [See Note (3)) Required Information and Notation (i) Hypothesized velue of the population proportion [Po] (LD) Actual value of the population proportion (a1 (iii) Lavell of significance [100(a)*] Gv) Power of the test [100(1-A)8) Exaggle 19 Recent surveys of vaccination coverage status in a large city have indicated 2 rate of about 908 of the eligible children having been fully vaceinaed. A timber of areas in the city are, however, suspected of having appreciably lower vaccination levels of about 508 or lower thus requiring special attention. What minimum sample sizes should be sampled in these areas, Gnd what threshold Value should be used, co test the hypothesis that the Vaccination levels are not more than 50%, at the 5% level of significance end a power of 908? Solution: (i) Hypothesized value of the population proportion 908 (Et) Acta value of the population proportion 508 Gi) Level of significance 38 (iv) Power of the test 908 From Table lub in the column headed "0.90" and the row headed "0.30" it is found that n-8 and d*-5, This means that a sauple of 8 children must be taken in each of the suspect areas and an area would be designated as not heving reached vaccination levels of over 50% if more than 5 children were found to be inadequately vaccinated. Sample size formila for Situation 1 n= (21altPo(d-Bo)) * en.p/tPa(-Pad!12/(PoPad? (3) and &* = [nP9-21-a/(nFo(1-Po))] {See Notes (11) ana (11i)] A full discussion of the LQAS method vill be availeble in the proposed revision of the WHO publication: "Adequacy of Sample Size’, The value of 4” 1s always rounded down (e.g. 5.3-35 5.85). "zy.g and "2,.g" represent the number of standard errors fron'the mean ané are functions of the confidence level and pover respectively, "a" and "8" refer to the type-I end type-IT Grxors respectively. These terms will be fully explained in the above mentioned publication,WHO/HST/ESH/86.1 Page 26 INCIDENGE RATE STUDY SITUATIONS 6.0, Estimation of Ineidenee Rates SITUATION 12: Estimation of incidence rates (Table 15 and Note on page 24) pigereine the minimm sample size required to estimate an incidence rate to within specified limita of ies true value Required Information and Notatton (4) Relative precession £ (1) Confidence level 100(1-a)e Ho large a sample of patients should be followed up in order to estinate the Imeidence rate of the disease within 108 of {te true vatae ean ese confidence? Solus: G) Relative precision 108 (1) Confidence level 958 pea pale 3S, at the interusction of the row corresponding to ¢ = 0.10 indteeten Tim SOFFepending to the 954 confidence level, a sanele oF 3852 indicated, Sample size formula for Situation 1: te size formils for Situation 12: 2 (tayo 8)? (16) [See Nore) "2-9/2" tepresents the number of standard errors from the mean fad is a function of the confidence level. This term will be fully explained in the proposed revision of the WHO publication: “Adequacy of Sample Size"WHO /ST/#94/86.1 Page 25 6.b. Testing Hypothesie for Incidence Rate SITUATION 13: Hypothesis testing for incidence rate (Table 16 and Notes on page 26) Minimum semple size for testing the hypothesis that the incidence rate of characteristic is equal to @ particular value Required Information and Notation Gi) Hypothesized value of the population incidence rate {Ao] (is) -Aetual value of the population incidence rate [Aa] Git) Level of significance |(100(a)8) (Gv) Power of the test (100(1-f)8) (@) Alternative hypothesis Ag > Ap Cor Aa < Ao? (one-sided test) or dat Xo (two-sided test) Example 21: ‘The incidence rate of a particular disease based on a 5-year follow-up of a small number of people is reported to be 40%, What minimum sample size Gould be needed to test the hypothesis that the incidence rate is 40e at the 5% level of significance? It is destred that the test should have @ power of 908 of detecting an incidence rate of 50% (i.e. rejecting the hypothesis) if in fact that were the true incidence rate Solutio: (i) Hypothesized incidence rate 408 (ii) Actual inefdence rate 508 Gat) Level of significance 58 (4x) Power of the test 90% (w) Alternative hypothesis imeldence rate greater than 40 From Table 16b (for the level of sigaificance of 58 and a power of 908) {see Note (4)) in tho column headed "0.40" and row headed "0.50" a figure of 169 is given a che minimum sample that would be needed. Sample size formlse for Situation 13: For one-sided test: = (Zea Ao + 2168 Aad?/onAa? For two-sided test: = (21-a/2 Not tLe had?/orAa)™ «sy [see Nore (41)]WHO/HSt/ESH/A6.1 Page 26 Notes: w This is a one-sided test since we are interested in rejecting the pull hypothesis only when the actual incidence rate 1s greater chan the hypothesized one. Tables giving other cholees of level of significance and power may be found in the propased revision ‘of the WHO publication: "Adequacy ef Sample Stee" GFl-a/2"s “Zia” and "21.9" represent the number of standard errors from the mean and are functions of the confidence level and power respectively. "a" represents the type-I error and 6 Feprosents the type-II error. These torms will be fully explained in the above mentioned publicationWHO/HST/ESM/86.1 Page 27 6.c. Sample Size for Testing Hypothesis for Two Incidence Rates SITUATION 14; Hypothesis testing for two incidence rates (lable 17 ana Notes on page 28) Minteum sample size for testing che hypothesis that the true incidence rates in two groups of individuals, with a comon date of entry in the study but no prior fixed end peint, are equal. [See Note (3)] Required Information and Notation (A) Rypothesized value of the difference between the population incidence rates [11-\9 ~ 0] (it) Aetual valuee of the Incidence rates (2 and Ag] (ii) Level of significance (100(a)8) (iv) Power of che rest [100(1-A)8] (~) — Alrernative hypothesis: Agoda > 0 Cor Ap-Ag <0) (for one-sided test), ApAg = 0 (for two-sided test) Example 22 As part of @ study of the long-term effect of noise to the workers in a pareicularly notsy industry, it is planned to follow-up a cohort of people who wore recruited into one such factory during a fixed period of time, and a binilar cohort of individuale in an dnnecuous industry, for the rest of their Lifetime or until their hearing is inpsired. How many people would be followed-up in each group to test the hypothesis that the incidence rates for bearing impairment in the two groups are the same, at the 5% level of significance with a pover of 80%? | The alternative hypothesis is that hearing impairment in the innocuous induscry is not more than the national average of about 208 whereas in the nofey industry it is 2.5 times (i.e. 508) Solution: (4) Hypothesized value of the difference: 0 (ii) ‘Incidence rates 508 and 20% Git) Level of aignificanc 58 (iv) Power of the test: 808 (@) Alternative hypothesis: ine{dence rates for the two groups different value of 23 is found in Table 17b as the entry in the column headed and row headed "0.20". [See Nove (14)] Semple size formulas for Situation 14 For L-sided test: f= (2p _ll HA?) #21 g/002 19222) 12/(a,-29)? For 2-sided test: a fetvq/2/ (LMA) + 2p -p/4d2yw129)12/(04-29)? (20) [See Notes (111) and (iv)]wHo/HSt /Es4/86 2 Page 28 In this situation subjects are inducted in the study and each followed-up until they either develop the characteristic under study or cannot be followed-up any more, stare of scuay where x = manifestation of characteristic ‘This is a two-sided test since we are interested in rejecting the PULL hypothesis irrespective of the direction of the difference between the rates of the two groups "k" is the number of study subjects in group 2 for each of the study subjects in group 1, In most studies k = 1 is used; hence the tables given here and che examples assume a one-to-one correspondence of study subjects in both groups. "21-a/2"; “21.q" and "21.9" represent the number of standard errors from the mean and are functions of the confidence level and power respectively, "a" represents the type-I error and & represents the type-I1 error. These terme will be fully explained {n the proposed revision of the WHO publication: "Adequacy of Sample Size". is the average of Ayand 2.WHO /HST/ESH/86.2 Page 29 SKTUATION 15: Ih dence rates in follow-up studies (iotes on page 30) Mintmun sample size for testing the hypothesis thac the true incidence rates in the two groups of subjects are equal when subjects axe followed-up until a comton date. [See Note (1)] Required Information and Notation (i) Hypothesized value of the difference between the population ineidence rates [21-42 = 01 Git) Actual values of the incidence rates [2 and 2] Git) Level of significance {100(a)¢] Gv) Power of the test [100(1-2)8] () Alternative hypothesis: Apa > 0 (or Ay-ag < 0) (for one-sided teat), or an #0 (Zor two-sided test). (v8) Duration of follow-up period: (T} [See Note (11)) Example 23 Suppose the study outlinad in Example 22 is to be undertaken with the subjects followed-up for 2 maxima of 5 yeara; what vould be the required sample size? Solution: [See Note (iv)] (4) —_-Hypothesized value of the difference: ° Gi) Inefdence rete 50% and 208 (His) Lavell of significance: 58 Gv) Power of the testi 808 (o) _-Alzernative hypothesie: Incidence rates different for two groups (vi) Duration of study: 5 years Using the formula for £(A) the intermediate computations give: £( = 0.35) = 0.2322 £1 = 0.50) = 0.3950 (a2 = 0.20) = 0.1087 And n= (1,96/(2(0.2321)} + 0.9416/(0.3950 + 0.2087)12/(0.5-0.2)? = 3.7351/0.09 = 41.5 ‘A sample of 42 would be needed for each group. {See Note (ifi)] Sample size formulae for Situation 15 For L-sided test: ne (pall (4K) £0)) + 2 pl WEA +£02) 172/01-22)? For 2-sided test: n= lap a/Wi re £0)] + zy pl{kEr)4£02)1)2/01-92)? (22) [seo Notes (it), (iv) and (~)]WHO/HST/ESH/86.1 Page 30 @ In this situation subjects are inducted into the study as they become available but wich a comuon end of follow-up period. Start end of of Study Study where x ~ manifestation of characteristic When study subjects are enrolled in the study throughout the period and stop the follow-up after time T, this controle the duration of the study but leads to censored observations. The hypothesized incidence rates have to be modified according to the formula: £(A) = a3 T/(MT-14e7 AT) No tables are given for this situation as the number of Parameters involved do not lend themselves to easy tabulations vk" fs the nunber of study subjects in group 2 for each of the study subjects in group 1, In most studies k= 1 &s used hence the tables given here and the examples assume a one-to-one correspondence of study subjects in both groups. "21-a/2") “Z1.q" and "21.g" represent the nunber of standard errors from the mean and are functions of the confidence level and pover respectively. "a" represents the type-I error and S represents the type-II error. These terms will be fully explained in the proposed revision of the WHO publication: “Adequacy of Sample Size". 1 is the average of \yand Az.WHO /HST/ES4/86,1 Page 32‘ble Ini Sample Size Necessary to Estimate P to Within q Feroentago Point with 96% confidence Antielpated Population Proportion (2) 0.40 0.980.300.2800 ons onto 70 oes 7203 ge? eases 205 207 ason_ ty Saas “ggg gm “es7, “aot ‘gana Sey 505 ast 55 Go? 2g oo seg 240 20g we says 2 sg ie 0 a ‘a =a @ & 3 2 67 gL S58 5D 2 8 63 oP © a@ » 2» 8 2 3 9% Rll 3 2 ee nz 2 op mw » ree ste a ae neg 2m 5 2% Y 2 ple 2 6 vy uw B Be oy 6 » bp ow wf wo oe Lb ob 3 u on pn % 8 ‘Table ab: Sample Sina Necessary to Estinate P to Within d Percentage Folnt Wish 908 Confidence Anticipated Population Preparticn (P) 245 0400.35 0.30 0.25 6597 tg ese sea isis lsat 3599 Naat 74s an 68h #9 4053856 260 260? we tng a7 og o> en? rr a 6 ar s& « f & o 8 3‘Table 2a: Sauple Size Necessary to Estimte P to within ¢ Fercentaga Boint with St Oonfidence Antieipeted Population Proportion (2) 0.05 0.30 0.15 0.20 0.25 0.30 9.25 G.€0 0.48 0.50 0. 0.65 0.70 0.75 0.80 0.88 0.90 729004 245744 217691 153666 125049 99658 71244 57624 46959 38416 31422 2561 20586 LE464 12806 9604 6780 4269 Yeza7e ‘seize seaz2 20416 2eni2 22420 17896 14606 11799 9604 7658 172 4116 3202 2401 1695 1068 LOL 3e4N 24198 17074 12806 9960 7928 e103 5217 A269 2499 2299 830 1423 1066 754 475 85615 21609 13605 9604 7203 3603 4459 3602 2935 2401 1965 3293 ozo sol ol az4 267 29197 19830 8708 41474610 3586 2854 7305 aye 537 1258 28 65923385 272 TTL ‘7299 4458 7177s? 1153897 Td 577 470385 TS 207 165 ze 96 88S 32M 1537 968 603mg 39937257 a0 71 AO 2 787 38 haz “be5 54538559 aks dak e869 2 2 24a 7 D lise 558 924685 SSK wD 3 oh a6 7 35242. oS Bw BL 5 see 283m 6k RG yo uon 4 #7 6 S95 2K 5G Dm] bo ou 3 3 ger YR e758 Ska n 9 7 3 mon 9 2 4% 3% 2 m 8 6 LD 2 7 6 2 ‘rable 2b: Sample Size Necessary to Batinate P to Within £ Reroentage Point with 90% Confidence anticipated Population Proportion (F) 0.05 0.10 0.15 0.20 0.25 0.90 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.00 0.85 0.50 ssxdoaa 243513 155229 109228 B127L 63159 50249 40586 39070 27067 72138 16032 14570 11596 OLS 6765 4775 3007 Bosal “eos79 "38331 27057 20299 as7ee 328e0 10147 8269 6765 565 4510 3643 2699 2255 1691 1198 752 Zrial 29057 17036 12026 9029 7015 SP 4510 3675 9007 2460 2005 ioie 1289 loo> 762 ST Szuai iszz0 e583 6765 S07E 3946 3141 2537 2067 1691 1294 1129 511 725 Sek 423 299 Sbsat ‘ovat 6193 43303247 2526 2010 1624 1323 10s) B86 702 AD 464 S617) IST Shel 243 tsa oes #22 692 503 406 321 ZL 22 MBL 146 16 91 GBB 2ees los) 68248) 2GL 28224 asAT at opm Sk M202 es “coo Seq an 203 8 ae on ws SGT ae; 390 oaé«74««13DsMsBLS 5S we 38k S22 im wl am 7 5 48 37 30 3 2 NY 1 iO 2 be we a 7 2 2 3 7 2 lm i wo So iss 9s) 68 SL 40S LT mom 7% ss 4 2 6 20 Vv M nh $ & 6 os 9826 DTD 65(Reble 3: Sample Size for One-Geaple Estimate Gavel of significance: $47 Rover: 901) Altarmative hypothesis: I-sidat) ‘Test Proportion Py ioe Fa 0205 0.10 0.25 0.20 0.25 0,30 0.25 0.40 0.45 0.50 0,55 0.60 0.65 0.70 0.75 0.80 0,85 0.90 O20 221 378 loo 33 22 16 we wy 8 6 os “67 “we uy 6 39 25 eb oe 0:20 38 @1 wi 75 i 2 2 to -25 21 49 131 seo ges 0 as ae ae OE 90 18 30 2 iss oe "es ae oe pO 935 11 30 36 72 176 746” 04 208 ge odo 8 ie 24 42 80 12 758 BY OB ods 7 a7 a7 46 a6 toa om de eB O82 3 8 13 19 20 4 “mn 20 as2 a AD ose Mom mm ‘93 a as Be 60s 2 52 93 21 o37 os 3 33 51 “52 205 sou 2.70 3 3 50 “a5 ise 2 2 2 i i BRET eoueven 2 ons 16 2 31 a 2.80. ub» Ss 4 2 4 @ B8seseeeesneus 0:05 how is we 2 ie wo 0:90 0.35 ble 4; Sample Size for One-Semple Test of Proportion (evel of aigniticance: 5%; Povart 9087 Alternative hypothmais: 2-eldad) ‘Test Proportion (2) 0.10 0.38 0.20 O25 37057 9906 4323 “5025, 19482056 ios lamp mB 835 ws 2k 87 “oy 5 56. 3 Se a 8 yo b oY 30 SeSRSebu5RRe8‘Table 6a: Values of V {Y= F(R) + Pp (2eP2)) Poe (2) 0.20 0.15 0.20 a aa BEESEEE SSSesRe SRSSEBESBERRBEEERBEERS REBRESEESEEEESSRaSReee BESRREEEREBERERSSESRRRSRE BUREDBESBRSRRRES geeng gees sesee BSRRERRERES! saarosbssse RBSRRERERSRERRGLEES BER SR ee BSSRSESREEREESEBEEEE SBESERRERRERBEEEES ‘table Sb: Sample Size to Hstinate the Rak Ditterence Between up Proportions ‘1 aie Bp to Within d Percentage Foints vith 96% Contidanee fusing V= Pye) + PGP) 8 8 REBemusamnneuunnn SEIQSRRERSA SSSR SSEE ESSE PSUURE ERE SCEREE ENS omsaaenune‘Teble 6: Sample Size for Two-smple Test of Proportion (level of significance: St; Rover! 90%) Alternative hypotheses 1-eided) A Py 0,05 0,10 0.35 0.10 478 788 018153743 2 217 990 53 109 273 3 @ 4s 73 Be 53 26 38 a 2 wo ue Fr‘Table @¢ Sample Size for Two-Gample Test of small Proportions (ievel of signiticance: Sty Power: 9047 Alternative Iypothesis: 1-elded) mL 9.0002 0.0002 0.000% 0.0005 0.0010 0.0025 0.0050 9.0002 2495sa ders eaa77 140063827135 9.0003 79092 Spen26 160602 20921 40051499 L000 42832 yeeoes 31678 «47521661 9:0008 28020 788085 ‘490559877 ocoon0 | “s15s ‘31679 49680 126532795 0002s 2674 ‘47925598 12658 3987 10089 1160 ies 1x7 27359947 0.0075 728 962 103 gl) ees. o1a100 527 866 Tor a2 7034885, 010200246 aan 300352, S07 a 9.0300 158 Wo ies L278 aL, L080 a7 8 Sw? 9.050092 in he Semple Size Necessary to Bytimate the Olds Ratio to within 10 Percentage Points with 95% Confidence as Ratio (28) a5 250.75 aos se49s 35032 29658 20060 anes ‘ere 4805 3908‘Table Sb: Semple size to Estimate the ols Ratio to Within 20 Peroartage Folnts with S58 Oandleanoe ots Ratio (oR) 9.25 0,80 0.75 1.001.251. 1,752.00 38500 20228 1132 15587 140s 3042 azaLe Tee you3 issn “oa 7873717 geld ase 584 aams 77961348303 480s 473 ann S31 50694825 O19 Get) 340908 7683 4723-9796. aa49 957 B9e2 Ded 3992 2405 ise ams sme asa aa 22 lel as) adore loa 476 1264 © 0639655 908s Ba Mase 202 “ee a2) 783753 9a 350099 7o7 795 a7 esta irr eck 7s 7D aka 92 7336s) gas GO ga 8 6a?) eka on 7 66 gs aT sa ma kaka 568 m6? eae 708 6 67m OL 57H 855 ml 820 a 65 a8 sae 1201376 asas as dae 2a SubaeBs B8s8RRh2‘Table di Gample Size Necessary to Betivate the ofMs fatio to within 50 Barcentage Points with 998 confidence (its Ratlo (08) SBSRBeSRSRBESRR: ‘Tabla 10: Sample Size for a Rypothacic Tost of the ois Ratio (St Level of wignificance ar SOb youu 2~tadl east) Oaks Ratio (08)wat Sample Size to Estinate the Relative Risk to Within 10 Rarcantage Points ‘with 99% confidence Relative Risk (RR) 0.25 0,80 am 1027 aseza 51217 56005 3304 9256525263, mee 20072 iseiz “3600 insu aa 79594499 62283462 50762769 22 2a 304 1903 3153 sis 27631384 2454 gs 292 8 80781 Mn 606 3061 ‘Tble Lib: Sample Size to mtimte the Relative Risk to Within 20 Percentage Polite ‘with 956 confidence Relative Risk (RR) BSSsasasseneeeaSample Sig to Eatinate the Ralative Risk to Within 25 Raroantage Points. with 95% Confidence Rolative Risk (FR) 0.25 0.80 0.78 1.50 2 pa? 13833 lore 7688 hsi2 “6670523 375 7634549357 2406 51093303 2es ia. 4926022073 ase 2228 300990 al Ms 85 aa 24 ass ue at D @ a 36 4 BRSSEESEERRS8E 1068603449 364k HO @. om 570305 72855 aa air a 16 2s 160 2S ms 16 37h ws ‘Table 11d: Sample Size to Bstimate the Relative Risk to within 50 Rerventage Points ‘with 95% confidence Rolativa Rick (RR) 0.25 0.50 goer 2303 ies hh a7 784 ‘903568 7a eh 34h 21 las aot lot as 80 wot OD ae no oR 7 8 B38 a OM a Beugnsaskessisarase‘Table 12: Sample Size for a Hypothesis Test of the Relative Risk (G4 level of significance are 30% powar 2-Gided tast) Relative Risk (FR) ‘Toble 23a: Minium Sample Sizos for tot Quality Asgurence Sampling No case acceptable with 95°% confidence EXEERERRESOS‘table 19b: Hinimm Suple Sizes for Lot Quality Asarance Sampling ‘Wo more than 1 case acceptable with $5 % confidence Prevalence ‘Table act wnimm Sample Sizes for Lot Quality Assurance Sampling io wore than 2 cases acceptahle with 99% contidenoe Prevalence‘Table 134: Minimum Sample Sizes fer Lot Quality Assurance Sampling No exe than 3 eases anceptable with 95 ¥ confidence ‘Table 130: Minimm Sample Sizes for Lot Quality Assurance Sexpling Te nore than 4" cases acceptable with 95 % confisence Prevalence SgesesesesuaTable 138: Minimm sample Sizas for Tot Quality Aesurance Sampling No case ecoeptable with 808 confidence Prevalence ‘Table 13g: Kinin Sample Stzos for Lot Quality Assurance sampling ‘Wo more than 1 case acceptable with 90 & confidence Prevalence BEREBEEEREBE‘Table 13h: Minimae Sample Sizes for Tot Quality Aswarance Sampling No more than 2 cases acceptable with 90 & confidence Prevalence BERBBEBEBERE BYSSEIORR LSE ‘Table 134: Minima Sample Sizes for Tot Quality Assurance Sampling Wo more than 3 cases acceptable with 90 ¥ eontidere Prevalence SEBBBBewwowe SSSEBESEBSEE BREBEERRE REE anagaasasess‘Table 12): Minimum Sample Sizes for Lot Quality Aeeuravee Seapling No more than 4 cases acceptable with 30 # confidence Prevalence BEBEBeecwwee Py u u % % u u uw u u Mu u ‘Teble 14a: Sample Size and Decision Rule for 108s (4 level of significance and 95¢ power, 1-sided alterative) Be (8) Beageet sone 265 1118 2077 151, SERKK ene BEeemeuure BBE ooe wn Sbeeeoxtoan 2 isu a 13 467 3409Table Léb: Sauple Size and Decision ile for TONS (6% Level of signizivcance ard 90% powur, Insided alternative) Fe (8) B 5 Ps 36 a 453 130 7 Sugie dies Tas Oar 5 Mc: Sample size and Decision Rule for 10k ‘Tobe fs (98 level of signiticarce ana 8ot power, Invided alterrative) Po (2) ‘ 10 8 15 a 2 6 36 ay 126 3 a ot 616 267 153 eu BREESE eee BREE veue % 0 18 80 8 BEaseeseou Bagkeasn Sanpie ala Tass a's‘Table 14d: Sample Size and Deciston Rule for TOAS (9% level of significance and S0b rover, 1-cided alternative) 6 10 1% 20 85 Berace Betea Broa. SEE Suaple Gis Teas than 5 Table 15: Sample ize to Hotimte the Incidwwoe Rata to within © per cent. with 98, 95t oF oot cenelaencs‘Table 16a: Sample Size tor Ona Sample Test of Incidence Density (levol of significance: Sty Power: 90¢7 Altarrutive hypothesis: 2-8ided) nm my Bn w 9 8 2 wo Bw KG RL wa mo 0 oe a is bb 2 zoe an ua $3 br 2 22 ub 7 7 9 2 38 GL 306 48 a7 a? 332 » 3 52 as 8 32 "83 Go” oe 88) a 2 2 45 lm 57579 a3 1s 72 3 6 29 61 160 7 | S70 378 37 a 1221 38 79 20 9a te? TL 63 eB to is 27 é& “99 252 ns6 ass 392 192 Is 33 2, 34 61 a2 306 1365 "2670 459 229, Ho 26 2 7% UT deca 294s Sat 20 14 aL 32 51 89 174 4502905 2242 32 1 25 38 61 104 204 500 2178, 3a B13 zt 30 48 7 aa 795 575 2400 8 30 1 1B 25 35 53 63 140 270 656 2825 3257 9 12 16 2 2 4 GL 95 160 306 741 3200 2 1 ld 18 24 33 47 69 0B Isl 345 32 3556 aeaucunase ol EP ‘Table 1b; Sauple Size for dhe Sample Tast of Incidence Density (level of signiticance: 54; Pover: 9Ct;" Altemative hypothesis: I-sided) hq 0.05 0,10 0.15 0.20 0.25 0.30 0.35 0.60 0.45 0150 0,55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 v7 2B we 7 7 6 ws 9 Mo 2 10 8 12 "me oa 6 21 265 1 9 aL 3 2s 368 aon 87 7 2774 358 ua 72 33 loz am eas a9 2 51 ss 788 32 6s 16 700 2 4 as 209 934 3B 29 51 102 254 5 ra a w 9 3 SBEREER Loe 2 © 12 6 Boo B % 38 2 aA 4 10 ae 25 ° ie 2 beevuueueueennnaed pe eeounoase Se vavaosseokke‘Tble 17a: Sample size for Test of Bouslity of meidence Densities (St level of significance aml Sot power sided altamative) se BiGBEesseeusre 88 ‘ble 17: Samle Size for Test of Byality of Incidence Densities (98 Level of significance and aot power 2-sided alternative) O10 O15 0420 0.25 0.30 0.35 0.40 0.45 0.59 0.55 0.60 0.65 0.70 0.75 0.90 0.05 7 7 2 2» 7 7 mw 7 ww aoe Bok 39 3 2 37 7 we 319 asp 11s 394 a2Teble 17cr Sample Size for Test of Bqality of Incidence Densities (28 level of significance and 30% povar z-sided alternative) a 0110 0,18 0420 0.28 0. 0.48 0.60 0.85 0.60 0.65 0.70 pw mo w “uu B Bw Bw us a2 Bb BY v we 2 mm 2 2B m 2 wD b 9 33m 2 Os Ee ee re a a re ey at uo 870 we 103 83 229 160 22 3264 903440 9 520 317 05 sel SS38UeErenaeween ‘table 17d: sample Size for Test of Equality of Tnofdence Denettins (4 Level of significance and 80% paver 2-aided alternative) 0.10 015 0.20 0.25 2 eas us eS 08