Additional digits (including zeros) should not be added, eg, after the decimal point, to provide all data entries with the same number of digits. Doing so may indicate more precise results than actually were calculated or measured. A percentage or decimal quotient should contain no more than the number of digits in the denominator. For example, the percentage for the proportion 9 of 28 should be reported as 32% (or decimal quotient 0.32), not 32.1% (or 0.321) (see 20.8, Statistics, Significant Digits and Rounding Numbers). Values reporting laboratory data should be provided and rounded, if appropriate, according to the number of digits that reflects the precision of the reported results to eliminate reporting results beyond the sensitivity of the procedure performed (see 18.4.1, Units of Measure, Use of Numerals With Units, Expressing Quantities).
Values for reporting statistical data, such as P values and confidence intervals, also should be presented and rounded appropriately (see 20.8, Study Design and Statistics, Significant Digits and Rounding Numbers). Although some publications2(p512) suggest use of specific designations for levels of significance (eg, a single asterisk in the table to denote values for entries for which P < .05, a dagger for P < .01), exact P values are preferred, regardless of statistical significance. In most cases, P values should be expressed to 2 digits to the right of the decimal point, unless the first 2 digits are zeros, in which case 3 digits to the right of the decimal place should be provided (eg, P = .002). P values less than .001 should be designated as “P < .001,” rather than using exact values, eg, P = .00006. For study outcomes, individual statistically significant values should not be expressed as “P < .05” either in the table or in the table footnote, and nonsignificant P values should not be expressed as “NS” (not significant). For confidence intervals, the number of digits should correspond to the number of digits in the point estimate. For instance, for an odds ratio reported as 2.45, the 95% confidence interval should be reported as 1.32 to 4.78, not as 1.322 to 4.784.