Abstract
Sample size calculation is an important element of research design that investigators need to consider in the planning stage of the study. Funding agencies and research review panels request a power analysis, for example, to determine the minimum number of subjects needed for an experiment to be informative. Calculating the right sample size is crucial to gaining accurate information and ensures that research resources are used efficiently and ethically. The simple question “How many subjects do I need?” does not always have a simple answer. Before calculating the sample size requirements, a researcher must address several aspects, such as purpose of the research (descriptive or comparative), type of samples (one or more groups), and data being collected (continuous or categorical). In this article, we describe some of the most frequent methods for calculating the sample size with examples from nuclear cardiology research, including for t tests, analysis of variance (ANOVA), non-parametric tests, correlation, Chi-squared tests, and survival analysis. For the ease of implementation, several examples are also illustrated via user-friendly free statistical software.
Similar content being viewed by others
Abbreviations
- TPD:
-
Total perfusion deficit
- SPECT MPI:
-
Single-photon emission computed tomography myocardial perfusion imaging
- IR:
-
Iterative reconstruction
- AC:
-
Attenuation correction
- RR:
-
Resolution recovery
- CMR:
-
Cardiac magnetic resonance
- CCT:
-
Cardiac computed tomography
- SSS:
-
Summed stress score
- RVEF:
-
Right ventricular volume and ejection fraction
References
G*Power: Statistical power analyses for Windows and Mac. Heinrich Heine Universität Düsseldorf, 1 Jan 2013. http://www.gpower.hhu.de/.
G*Power data analysis examples: Power analysis for paired sample t-test. UCLA Institute for Digital Research and Education. http://www.ats.ucla.edu/stat/gpower/pairedsample.htm.
Chow SC, Shao J, Wang H. Sample size calculations in clinical research. 2nd ed. Boca Raton, FL: Chapman & Hall/CRC; 2008.
Julious SA. Sample sizes for clinical trials. Boca Raton, FL: Chapman & Hall/CRC; 2010.
Qutub MA, Dowsley T, Ali I, Wells RG, Chen L, Ruddy TD, et al. Incremental diagnostic benefit of resolution recovery software in patients with equivocal myocardial perfusion single-photon emission computed tomography (SPECT). J Nucl Cardiol 2013;20:545-52.
International conference on harmonization of technical requirements for registration of pharmaceutical for human use. Statistical principles for clinical trials E9, Feb 1998. http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Efficacy/E9/Step4/E9_Guideline.pdf.
Devroye L. Non-uniform random variate generation. New York: Springer; 1986.
Machin D, Campbell M, Fayers P, Pinol A. Sample size tables for clinical studies. 2nd ed. Malden: Blackwell Science; 1997.
Venkataraman R, Wael A, Belardinelli L, Heo J, Iskandrian AE. The effect of ranolazine on the vasodilator-induced myocardial perfusion abnormality. J Nucl Cardiol 2011;18:456-62.
Desu MM, Raghavarao D. Sample size methodology. New York: Academic Press; 1990.
Stuart A, Ord JK. Kendall’s advanced theory of statistics. 6th ed. London: Edward Arnold; 1994.
Draper NR, Smith H. Applied regression analysis. 3rd ed. New York: Wiley; 1998.
Kendall M, Gibbons JD. Rank correlation methods. 5th ed. New York: Oxford University Press; 1990.
Wang L, Zhang Y, Yan C, He J, Xiong C, Zhao S, et al. Evaluation of right ventricular volume and ejection fraction by gated 18F-FDG PET in patients with pulmonary hypertension: Comparison with cardiac MRI and CT. J Nucl Cardiol 2013;20:242-52.
Wild CJ, Seber GAF. Chance encounters: A first course in data analysis and inference. New York: Wiley; 2000.
Doukky R, Frogge N, Bayissa YA, Balakrishnan G, Skelton JM, Confer K, et al. The prognostic value of transient ischemic dilatation with otherwise normal SPECT myocardial perfusion imaging: A cautionary note in patients with diabetes and coronary artery disease. J Nucl Cardiol 2013;20:774-84.
Fleiss JL, Tytun A, Ury HK. A simple approximation for calculating sample sizes for comparing independent proportions. Biometrics 1980;36:343-6.
Cohen J. Statistical power analysis for the behavioral sciences. 2nd ed. Hillsdale, NJ: Lawrence Erlbaum Associates; 1998.
Collett D. Modeling survival data in medical research. 2nd ed. London: Chapman & Hall/CRC; 2003.
Abraham A, Nichol G, Williams KA, Guo A, deKemp RA, Garrard L, et al. 18F-FDG PET imaging of myocardial viability in an experienced center with access to 18F-FDG and integration with clinical management teams: The Ottawa-FIVE substudy of the PARR 2 trial. J Nucl Med 2010;51:567-74.
Cheung YK. Sample size formulae for the Bayesian continual reassessment method. Clin Trials 2013;10:852-61.
Everitt BS. Repeated measures analysis of variance. Wiley 2014, StatsRef: Statistics Reference Online.
Disclosures
Dr Einstein has received institutional research grants for other investigator-initiated studies from GE Healthcare, Philips Healthcare, Spectrum Dynamics, and Toshiba America Medical Systems. The authors declare no other conflicts of interest.
Author information
Authors and Affiliations
Corresponding author
Additional information
See related editorial, doi:10.1007/s12350-015-0288-z.
Appendix: G*Power Software
Appendix: G*Power Software
Case Study for Two-Sample Independent t Test (Two-Tailed)
For the two-sided, two-sample independent t test, select the appropriate test under the Test Family dropdown (see Fig. 2). The statistical test used will be selected from the dropdown as “Means: Difference between two independent means (two groups).” For all the analyses in this Appendix, we will select “A priori: Compute required sample size—given α, power, and effect size” since we are calculating sample size. Ensure that “Two” is selected for Tails. From here, the first step we would like to calculate is the effect size d. We can do this by clicking the “Determine” button to open the extended window. Where n1 = n2, input the means and standard deviations for each group and click “Calculate and transfer to main window” (Note: The means for groups 1 and 2 were adjusted so their difference is 1, as that is the clinically meaningful difference being assessed). We then include our standard type I error and power of 0.05 and 0.80, respectively, and leave the Allocation Ratio as 1 (we are assuming equal allocation). From here, clicking “Calculate” will yield the approximate sample size, similar to that which we calculated manually.
Case Study for Two-Sample Independent t Test (One-Tailed)
For the two-sample independent t test using one tail, select the appropriate test under the Test Family dropdown (Fig. 3). The statistical test used will be selected from the dropdown as “Means: Difference between two independent means (two groups).” Ensure that “One” is selected for Tails. From here, the first step we would like to calculate is the effect size d. We can do this by clicking the “Determine” button to open the extended window. Where n1 = n2, input the means and standard deviations for each group and click “Calculate and transfer to main window” (Note: The means for groups 1 and 2 were adjusted so their difference is 1, as that is the clinically meaningful difference being assessed). We then include our standard type I error and power of 0.05 and 0.80, respectively, and leave the Allocation Ratio as 1 (we are assuming equal allocation). From here, clicking “Calculate” will yield the approximate sample size, similar to that which we calculated manually.
Wilcoxon Rank-Sum Test (Non-parametric Version of Two-Sample Independent t Test)
For the Wilcoxon Rank-Sum Test, select the appropriate test under the Test Family dropdown (see Fig. 4). In this case, we are using the t tests still. The statistical test used will be selected from the dropdown as “Means: Wilcoxon-Mann-Whitney test (two groups).” Ensure that “Two” is selected for Tails. From here, the first step we would like to calculate is the effect size d. We can do this by clicking the “Determine” button to open the extended window. Where n1 = n2, input the means and standard deviations for each group and click “Calculate and transfer to main window.” We then include our standard type I error and power of 0.05 and 0.80, respectively, and leave the Allocation Ratio as 1 (we are assuming equal allocation). From here, clicking “Calculate” will yield the approximate sample size using Wilcoxon simulations.
Case Study for Two-Sample Proportions
For two-sample proportions, select the appropriate test under the Test Family dropdown. In this case, we are using the z test (see Fig. 5). The statistical test used will be selected from the dropdown as “Proportions: Difference between two independent proportions.” From here, we input our two proportions under Proportion p2 and Proportion p1. We then include our typical type I error and power of 0.05 and 0.80, respectively. The allocation ratio we are looking at is 1, as we would like for our groups to have equal sizes. From here, clicking “Calculate” will yield the approximate sample size, similar to that which we calculated manually.
Fisher’s Exact Test
For two-sample proportions where the proportions are assumed to be different between exercise and vasodilator stress, select the appropriate test under the Test Family dropdown (see Fig. 6). In this case, we are using the exact test. The statistical test used will be selected from the dropdown as “Proportions: Inequality, two independent groups (Fisher’s exact test).” From here, we input our two proportions under Proportion p1 and Proportion p2. We then include our typical type I error and power of 0.05 and 0.80, respectively. The allocation ratio we are looking at is 1, as we would like for our groups to have equal sizes. From here, clicking “Calculate” will yield the approximate sample size.
Rights and permissions
About this article
Cite this article
Chiuzan, C., West, E.A., Duong, J. et al. Sample size considerations for clinical research studies in nuclear cardiology. J. Nucl. Cardiol. 22, 1300–1313 (2015). https://doi.org/10.1007/s12350-015-0256-7
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12350-015-0256-7