Comment on ‘Testing for ALK rearrangement in lung adenocarcinoma: a multicenter comparison of immunohistochemistry and fluorescent in situ hybridization’

To the Editor: I read with great interest the recent study performed by Selinger et al.¹ exploring the potential utility of immunohistochemical screening for ALK rearrangements. Although the results of the study are convincingly portrayed, I did note some potential omissions worth pointing out, especially given the major impact that the study might have on the logistical approach that many laboratories might choose to take as part of the very important work-up of non-small-cell lung cancers (NSCLCs).

Although the methods outline the origin of the sample, the authors do not elaborate upon any form of a priori sample size calculations. These calculations can be easily performed using web-based tools; alternatively, a number of medically oriented reviews are available explaining the calculations involved. Using the worked examples of Jones et al.,² given the assumption of ALK rearrangement in 3–4% of cases of NSCLC,¹ and presuming test sensitivity of at least 95% (which is often assumed in sample size calculations for studies of screening tests²), I calculated a minimum sample size of 2086. This suggested that sample size is much larger than that used by Selinger et al.¹ and relates to the very low prevalence of ALK rearrangements in NSCLC.

Another vitally important omission is the complete lack of any statements relating to the statistical quantification of uncertainty (eg, 95% confidence intervals); this violates one of the major tenets of diagnostic testing research, as set out by the STARD group.³ Admittedly, the lack of any false-negative immunohistochemical results in this study makes accurate calculation of confidence intervals difficult. Nevertheless, estimates are possible: if one uses the online tool provided by the KT-clearinghouse group⁴ (with the caveat that a 0 false-negative value be estimated by a small non-zero value, eg, 0.01), the sensitivity in the current study can be estimated at 100% with 95% confidence interval ranging from as low as 64 to 100%. This wide confidence interval also likely relates to undersampling.

A final comment highlighting the importance of follow-up FISH studies pursuant to a positive ALK immunohistochemical result is warranted. Such a statement is an important one in order to ensure that laboratories hoping to optimize their efficiencies not assume that a positive ALK immunohistochemical test equates to a positive ALK rearrangement (especially in light of the very low positive predictive value noted in this study).