Expertise or ideology? A response to Morris et al. 2016, ‘Circumcision is a primary preventive against HIV infection: Critique of a contrary meta-regression analysis by Van Howe’

Click to increase image sizeClick to decrease image size

Notes

1 In a 2016 letter to the editor in response to a study (Bossio, Pukall, & Steele, 2016) confirming the findings reported by Sorrells et al. (2007), Morris and Krieger continue to demonstrate their ignorance of marginal mixed models and the proper application of the Bonferroni adjustment. They continue to insist that the Bonferroni adjustment should be applied to the data reported by Sorrell and colleagues (2007) although they cite a clear explanation of why the adjustment is only appropriate when multiple comparisons are being made or multiple hypotheses are being tested (Van Howe, 2014). They also expressed incredulity over the inclusion Hispanic ethnicity and type of underwear worn in the multivariate analysis. A marginal mixed model incorporating multiple measurements from each participant made one comparison to test a single hypothesis that had been developed using pilot data, so it would inappropriate to apply the Bonferroni adjustment. There is an association between ethnicity and circumcision status that is potentially confounding and the type of underwear worn may also impact penile sensitivity. Someone with primary source expertise would know that Bonferroni adjustments are not used with marginal mixed models (Brown & Prescott, 1999; Littell, Millikein, Stroup, Wolfinger, & Schabenberger, 2006; Verbeke & Molenberghs, 2000; West, Welch, & Galecki, 2014) and understand the importance of identifying and evaluating potentially confounding factors (Mendenhall & Sincich, 2011).

2 While past rebuttals from these authors have revealed an ignorance of basic statistics and mathematics, let alone more complicated statistical models (Svoboda & Van Howe, 2014; Young, 2007), and past performance is the strongest predictor of future performance, past transgressions will not be a part of evaluating the current critique.

3 Odds ratios derived from raw data were used in the meta-analysis (Tian et al., 2013) that one of the authors (BJM) took credit for (Morris, Bailis & Wiswell, 2014b), the meta-analysis one of the authors (BJM) was involved in (Morris & Wiswell, 2013), and the Cochrane systematic review on studies of HIV infection and male circumcision (Siegfried et al., 2003).

4 For example, Bossio et al. (2015) describe a review by Morris and Krieger (2013) as being of ‘low quality’ on account of the ‘high risk of bias … introduced by the authors’ well-documented, unconditional support of the practice of circumcision’ (p. 1306).

5 Similar search techniques were employed by two of the authors (BMJ, JNK) in systematic reviews that they have authored (Morris & Krieger, 2013; Morris & Wiswell, 2013).

6 As Petitti (2000) notes,

Published reports of studies that meet all the eligibility criteria for a meta-analysis may not present an estimate of effect size, and the raw data that would allow an estimate to be calculated may not be available in the study report. These studies cannot contribute to the summary estimate of effect. (p. 87)

7 For example, in the 2003 Cochrane review only 4 of the 34 studies available at that time were case-control studies (Siegfried et al., 2003).

8 In regression analysis, the terms effect modifier, interaction, and multiplier all refer to the same thing. In a regression model with two variables, the most common model evaluated is:When evaluating for interaction between x₁ and x₂, one must add a term that multiplies the two variables (x₁ × x₂), then evaluate whether the parameter β₃ is statistically significant using the equation:

9 For these authors, it is not uncommon for a cited reference to fail to support the assertion made in the statement cited. In the critique, the authors mischaracterize the findings of Bärnighausen and colleagues (2012).

10 Generally, see Collier (2012). The Clinton Health Access Initiative has sponsored studies assessing the effectiveness of the roll-out of circumcision in Africa (Vandament et al., 2016). PEPFAR is a sponsor of the circumcision roll-out in Africa and studies assessing the effectiveness of the roll-out (Osaka et al., 2015) and has funded studies promoting infant circumcision in Africa (Amuri et al., 2016; Fitzgerald et al., 2016). The Bill and Melinda Gates Foundation also promotes infant circumcision in Africa, (for example, Mavhu et al., 2016; Mangenah et al., 2016) and funded a trial of circumcision in Africa (Wawer et al., 2009) that was clearly unethical (Berer, 2009; Gupta & Goel, 2009; Okwuosa, 2009). A member of the Editorial Board for Global Health: Science and Practice represents the Gates Foundation (http://www.ghspjournal.org/site/misc/edboard.xhtml). There is no evidence of an association between infant circumcision and the risk of HIV infection.

11 Gish gallop refers to the creation of so much unsubstantiated material, at such a pace, that those who are committed to a more careful scholarly investigation do not have the time nor energy to rebut each point.