More Review Errors Shrink Evidence Base for HER2 Prognostic Role

HER2 is widely, even universally recognized as prognostic of adverse clinical outcomes in breast cancer. However, two review papers supporting this belief contain a remarkable number of errors, raising the question of what evidence now supports a prognostic role for HER2.

Correcting the errors in a 1998 review of 47 studies by Jeffrey Ross and Jonathan Fletcher overturns the review's conclusion that HER2 is independently prognostic. Ross did not dispute the corrections.

The 47 papers and the errors of the 1998 review are included in a 2003 update from Ross et al. The 2003 edition adds 34 more papers and introduces 10 new errors. All told, the 2003 review examined 81 papers and erred on 20. 

I previously documented the mistakes of the 1998 review. There were nine coding errors and two papers that should not have been included in the review. (One of the two papers was also miscoded, but I only count the paper mistaken once, making for 10 total errors rather than 11.)

The 2003 review adds the following 10 new errors:

  • 5 papers coded 'Yes' for multivariate significance should be 'No'
  • 2 papers coded 'Yes' for multivariate significance should be 'NA' 
  • 1 paper should not have been included 
  • 2 papers coded 'No' for multivariate significance should be 'Yes' 

The basis for these conclusions are found in Appendix I below.

Contacted regarding these errors, first author Jeffrey Ross replied that because he was traveling, he didn't "have complete access to review your findings." But, continued Ross: "I have no reason to believe that your conclusions are not correct and that there were scattered errors in the meta-analysis of the published literature in our 2003 manuscript."

The scope and scale of the errors might make both papers candidates for correction or retraction. The Oncologist published both. Executive Editor Martin Murphy did not reply to an email regarding problems with the 1998 review.


Appendix I

5 papers coded 'Yes' for multivariate significance should be 'No'

1) Jukkola et al. (2001)

The abstract reports: "In multivariate regression analysis, only tumour size and nodal involvement were risk factors for poor survival when analysed separately together with c-erbB-2 and receptor status..." 

Section 3.2 states: "In multivariate Cox stepwise regression analsis, tumour size and nodal involvement emerged as independent prognostic factors when analysed separately in combination c-erbB-2, indicating a 2.9 (90% CI 1.9-4.4) risk of death in node-positive patients. For patients with tumour sizes T3 or T4 the risk of death was 2.7 (90% CI 1.4-5.1) and 4.8 (90% CI 2.5-9.5), respectively, c-erbB-2 status did not reach significance in this model, nor when analysed in combination with tumour size, nodal involvement and receptors."

2) Rudolph et al. (2001)

HER2 only emerges as prognostic if CR is removed: "When all variables that attained statistical significance in the univariate analysis were included in the multivariate model, the CR was the first and most significant independent indicator of both AOS and DFS (P  .0001; Table 3). Next to CR, only PR status was found to be an independent prognostic factor, albeit of borderline significance."

3) Pinto et al. (2001)

HER2 is not independently prognostic: "C-erbB-2 is an independent prognostic indicator when evaluated in conjunction with ploidy and SPF." 

4) Suo et al. (2002)

HER2 is only prognostic when combined with EGFR or HER4. See Table 5. 

5) Spizzo et al. (2002)

The paper states: "Multivariate analysis for DROS revealed that nodal status, EpCAM overexpression, tumor size and histological grade were significant prognostic factors. Hormone receptor expression and Her-2/neu overexpression were not significant predictors of DROS. For DFS, nodal status, Ep-CAM overexpression, tumor size and progesterone receptor expression were significant prognostic factors. Her-2/neu overexpression, histologic grade and estrogen receptor expression had no prognostic value for disease-free survival (Table III)."

2 papers coded 'Yes' for multivariate significance should be 'NA' 

1) Agrup et al. (2000)

No multivariate analysis

2) Horita et al. (2001)

No multivariate analysis

1 paper should not have been included 

Wright et al. (1989) is one of three studies incorporated in Gullick et al. (1991) with the result that the same 185 patients are counted twice. 

2 papers coded 'No' for multivariate significance should be 'Yes' 

1) Scorilas et al. (1999) 

Tables 2 and 3 show HER2 overexpression prognostic in multivariate analyses of early relapse and overall survival.

2) Rosenthal et al. (2002) 

A paper on which Ross is senior author found "Multivariate analysis of the combined LN+ and LN− lobular and ductal cases revealed that HER-2/neu amplification (P   0.002), pathologic stage (P < 0.0001), and node positivity (P < 0.0001) were all independent predictors of disease-related death."