The identification of MMR mutation carriers is relevant, as there are individuals at high risk of developing cancer and can benefit from follow-up recommendations for the early detection of cancer. However, the cost of mutation detection by DNA sequencing is high, which creates the need for adopting strategies in order to reduce cost but maintain effectiveness. Although microsatellite instability (MSI) and immunohistochemistry have been incorporated into clinical practice (despite the fact that MSI is not a test widely available in Brazil) risk prediction models can be valuable tools that could be integrated in clinical practice.
According to the literature, this study presents a significant sample of individuals from South America [1
]. The number of mutations identified in the MSH2 gene was similar to the number of MLH1 mutations found. It was also observed that the most frequent extracolonic tumor in probands was endometrial tumor followed by breast cancer.
Weitzel et al. [42
] have pointed out risk prediction model applications that could be perfectly adapted in this situation: risk prediction models could help in the elaboration of reports to health insurance companies in order to get approvals for genetic testing; to provide realistic expectations to the patient regarding a positive result and to reinforce the absence of indication for genetic testing when a low probability mutation probability is calculated together with other screening techniques, such as microsatellite instability. However, for their use in the clinical practice, it is necessary that accuracy and predictive ability is thoroughly evaluated, since they may influence not only the adoption of a particular model but also its threshold.
In choosing the model for use in clinical practice three main points should be considered: the availability of the model; practical aspects in the management of the data; and the model performance.
All models are available via the Internet or through free software. When considering the advantages against time, the PREMM and Barnetson models are the best choice since they demand less time to be filled out. Conversely, MMRpro and Wijnen are available at the genetic counselling package CancerGene (CaGene) and a family's pedigree must be built, which is time consuming. In addition, the information available on CaGene must be stored and retrieved if novel elements from family history appear, demanding recalculations. As such, the structure and availability of human resources also influence the choice of the model.
Regarding the analysis of accuracy, all models used in this series presented AUC superior to 0.5. The largest AUC was from the Barnetson model, but this difference was not significant when compared to PREMM, MMRpro and Wijnen. Both Barnetson and MMRpro use information based on microsatellite instability and IHQ data, which can increase accuracy. In this study, since tumor samples of all subjects had not been taken, it was opted not to include this information. The Myriad model presented an AUC inferior to the four other models, a result also noticed by Monzon et al [28
Since there is no consensus in the literature about which threshold should be used in these models in order to better indicate molecular investigation, sensitivity and specificity of the five models were calculated according to ≥ 5%, ≥ 10%, ≥ 20% and ≥ 30% threshold. What is often observed is a variation in sensitivity and specificity according to the threshold and model used, which suggests that the use of a single threshold for all models (e.g., 10%) implies the alteration of both sensitivity and specificity which could lead to a different detection mutation rate. With the same threshold and similar sensitivities, as it occurs with a 10% threshold and 0.90 sensitivity by the Barnetson and PREMM models, the specificities are 0.33 and 0.58 respectively. When we consider threshold of 10%, the best relationship between sensitivity and specificity among models is regarding the PREMM model (sensitivity 90% specificity 54%).
As a consequence the healthcare professional must consider the characteristics and the performance of each model. The characteristics of each model should also be discussed and considered prior implementation in a genetic counselling practice.
The cost-effective analysis regarding Lynch Syndrome in South America is a point of debate. These studies are still at the initial stages. The evaluation of the costs will be of crucial importance for the implementation of public policies for genetic testing and management of risk individuals.
Considering the characteristics associated with a pathogenic mutation in our sample, the histologic type was the only that were not included in the models evaluated. It should be considered that this characteristic is not always available at the time for genetic counselling, therefore its inclusion could generate incomplete data.
The Brazilian population is extremely heterogeneous, the result of five centuries of the integration of individuals from three continents: Europeans, Africans and Amerindians [43
]. After colonization by the Portuguese, Brazil received a significant number of Africans. With the end of slavery, Brazil received European immigrant groups, which contributed to a strong European influence in the genome of the Brazilian citizen. According to Pena and colleagues [43
], genetic variation among Brazilians is so broad, that it should not be considered as a group but at the individual level.
Our study has limitations that should be considered. The molecular investigation was limited to direct sequencing, and other techniques were not used for molecular evaluation, such as MLPA, which could have increased the number of identified mutations.