Inaccurate data in meta-analysis; ‘A posteriori dietary patterns and metabolic syndrome in adults: a systematic review and meta-analysis of observational studies’

Madam

We read with interest the meta-analysis by Shab-Bidar et al.⁽¹⁾ on the association of a posteriori dietary patterns (DP) and metabolic syndrome. We noticed several inaccuracies regarding the inclusion of data that need to be clarified. The authors declare that the meta-analysis was conducted with articles published up to July 2015, but many articles^(2–6) meeting the inclusion criteria were not selected and included in the study. We found that the food composition of the DP included in the meta-analysis not always reflected the frequency of consumed foods characterizing the categories ‘Unhealthy/Western’ and ‘Healthy/Prudent’ investigated by the authors. They classified and analysed together as ‘Healthy/Prudent’, DP whose composition differed considerably, as described in the studies by Suliga et al.⁽⁷⁾, Akter et al.⁽⁸⁾, Hong et al.⁽⁹⁾, Kim and Jo⁽¹⁰⁾ and DiBello et al.⁽¹¹⁾. Similarly, for the analysis of the ‘Unhealthy/Western’ pattern, they combined different DP described in the studies by Gadgil et al.⁽¹²⁾, Hong et al.⁽⁹⁾, Kim and Jo⁽¹⁰⁾, DiBello et al.⁽¹¹⁾, Noel et al.⁽¹³⁾ and Esmaillzadeh et al.⁽¹⁴⁾. Therefore, the combination of these risk estimations seems methodologically incorrect. We summarize the misclassified DP included in the ‘Unhealthy/Western’ and ‘Healthy/Prudent’ patterns in Table 1.

Table 1 Summary and composition of the misclassified dietary patterns (DP) included in the meta-analysis

In addition, we noticed some inconsistencies regarding the risk data included in the meta-analysis. The most evident inaccuracies in the risk data for the ‘Unhealthy/Western’ pattern are the following: (i) study by Gadgil et al.⁽¹²⁾, ‘Animal protein’ DP, OR = 0·69 (95 % CI 0·65, 1·10) instead of OR = 0·69 (95 % CI 0·43, 1·10); and (ii) study by Suliga et al.⁽⁷⁾, ‘Fat, meat and alcohol’ DP, OR = 1·22 (95 % CI 0·97, 1·53) instead of the risk estimate of the adjusted model OR = 1·04 (95 % CI 0·82, 1·33). The inaccuracies in the risk data for ‘Healthy/Prudent’ pattern regard the following risk estimations: (i) study by Gadgil et al.⁽¹²⁾, ‘Fruits, vegetables, nuts, and legumes’ DP, OR = 0·80 (95 % CI 0·62, 1·51) instead of OR = 0·65 (95 % CI 0·38, 1·11); (ii) study by Suliga et al.⁽⁷⁾, ‘Healthy’ DP, OR = 0·68 (95 % CI 0·53, 0·92) instead of the adjusted model OR = 0·87 (95 % CI 0·68, 1·13); (iii) study by Naja et al.⁽¹⁵⁾, ‘Traditional Lebanese’ DP, OR = 1·96 (95 % CI 0·85, 4·51) instead of OR = 1·96 (95 % CI 0·82, 4·34); and (iv) study by Cho et al.⁽¹⁶⁾, ‘Healthy’ DP, OR = 0·58 (95 % CI 0·43, 0·78) instead of OR = 0·58 (95 % CI 0·50, 0·91).

In summary, since the dietary patterns represent a complex variable reflecting specific combination of different foods which varies consistently among the studies, we believe that pooling dietary patterns on the basis of factor loadings and combining risk data referring to similar dietary patterns are essential to obtain consistent and solid evidence on the association between diet and health-related outcomes as expected in a meta-analysis.