Statistical & Clinical Analysis
Diagnostic meta-analysis: an illustrative ESS-weighted Deeks regression slope
Calculate Deeks effective sample size and an illustrative weighted regression slope; a slope alone is not a publication-bias test.
Separate a weighted slope from a full asymmetry test
In diagnostic test accuracy reviews, effective sample size ESS = 4*n1*n2/(n1+n2) captures the imbalance between the diseased and non-diseased groups. Deeks and colleagues studied regression of log diagnostic odds ratios against 1/SQRT(ESS), weighted by ESS.
The five-study data below are fictional. With these inputs the ESS-weighted slope is approximately 16.85; an ordinary unweighted SLOPE returns 16.95 and is a different calculation. A slope alone is not a Deeks test, a p-value or evidence that publication bias exists. Interpretation requires an appropriate statistical analysis and consideration of heterogeneity and the small number of studies.
| Illustrative case | Study 1 ESS | Weighted slope | Interpretation |
|---|---|---|---|
| Five fictional studies | 75 | 16.85 | Descriptive slope only; no significance test |
| Constant log DOR | 75 | 0.00 | A zero slope alone does not establish absence of bias |
Calculate the ESS-weighted regression slope
E2:E6 contains ESS weights, F2:F6 contains 1/SQRT(ESS), and D2:D6 contains log diagnostic odds ratios. The formula rejects missing, nonnumeric, nonpositive or fractional group counts, and reports an undefined slope when all predictor values coincide.
=IF(COUNT(B2:D6)<>15,"Invalid study inputs",IF(OR(MIN(B2:C6)<=0,SUMPRODUCT(--(MOD(B2:C6,1)<>0))>0),"Arm sample sizes must be positive whole numbers",LET(sw,SUM(E2:E6),sx,SUMPRODUCT(E2:E6,F2:F6),sy,SUMPRODUCT(E2:E6,D2:D6),sxx,SUMPRODUCT(E2:E6,F2:F6,F2:F6),sxy,SUMPRODUCT(E2:E6,F2:F6,D2:D6),den,sxx-sx*sx/sw,IF(ABS(den)<0.000000000001,"Slope undefined",ROUND((sxy-sx*sy/sw)/den,2)))))Effective sample size for the first study
For 25 diseased participants and 75 non-diseased participants, ESS is 75. This quantity is not the number of participants and does not replace study-quality or bias assessment.
=IF(COUNT(B2:C2)<>2,"Invalid study inputs",IF(OR(B2<=0,C2<=0,B2<>INT(B2),C2<>INT(C2)),"Arm sample sizes must be positive whole numbers",ROUND(4*B2*C2/(B2+C2),2)))Try the example
Use the original values or change them to test your understanding.