الفهرس 
TitleOnly 14 pages are availabe for public view
 |  | from | 123 |  |  |
| | | from | 123 | | |
Abstract
Breast cancer (BC) is one of the major health problems facing women worldwide. It is the most common cancer in women representing 25.1% of all cancers and the second leading cause of cancer death after lung cancer.
Incidence and mortality in low-resource countries including Egypt have been increasing compared to developed countries due to the economic and social changes increasing risk factors as delayed pregnancy, having fewer children, shorter or no periods of breastfeeding, physical inactivity, greater levels of obesity besides the increase in BC awareness and screening .
The most serious issue is that most cases are already diagnosed at advanced or terminal stages. So, treatment efficacy is usually not satisfactory due to rapid growth, high metastasis leading to high rates of mortality. Therefore, screening and early detection strategies become mandatory. A variety of screening modalities are widely used namely mammography, ultrasound, Breast Self-Examination, and Clinical Breast Examination. The last two methods are inaccurate while mammography which is the best screening method faces many barriers; Most notably accessibility in only main governmental and private hospitals.
BC screening models have been developed as a need for quick, easy, safe as well as effective screening tool. Simply, they are statistical tools for estimating the chance of a woman with certain risk factors to develop BC within a certain time period (as within 5 years, 10 years or lifetime).
Two of these models; the Gail model being the standard risk evaluation model, and the Tyrer-Cuzick model being less widely applicable but more detailed including extensive family history added to personal factors. A high risk woman having Gail 5-year risk ≥ 1.67% or Tyrer lifetime risk ≥20% is recommended for annual MRI screening besides mammogram and other preventive measures.
In the present study, both models were tested for their accuracy among Egyptian population so as to limit and optimize the use of mammogram for high risk women identified by these models, getting the double advantage of minimizing both cost and X-ray exposure hazards.
Aims of the study
The present study aims to optimize the use of mammogram through:
Calculating BC risk among Egyptian women using the Gail and Tyrer- Cuzick software models.
Determining a cut-off value for positivity for the two models.
Evaluating the performance of the two models in BC screening.
Comparing the performance of the two models in BC screening
Summary, Conclusion and Recommendations
79
Materials and Methods:
a- Study Design:
A cross-sectional study was conducted.
b- Study population:
Females presenting for breast cancer screening.
c- Study settings:
Oncology units or centers conducting The National Egyptian Survey for breast cancer (President SISI Campaign for Egyptian woman support) during the period 2020- 2021.
d- Exclusion criteria:
o Females who refused to participate.
o Females below 35 or above 79 years old (as these ages are not accepted by the models).
o Females with history of pre-invasive or invasive breast cancer
o Females with history of breast surgery.
o Females excluded by the models (eligibility to use the model)
e- Sample size Calculation
The study population was sampled until a minimum sample size of (134) females with positive mammogram confirmed by positive biopsy and another (134) females with negative mammogram is reached, which is sufficient to estimate the accuracy of models based on an AUC of 0.66 with 90% power, and 0.05 level of significance using MedCalc software version 12.4.0.
f- Sampling technique:
The study population was sampled by the simple random method until the required sample size was reached.
g- Data collection
A data collection form was prepared representing both the Gail and Tyrer-Cuzick models. Females were interviewed before undergoing mammographic screening. After screening, Females were classified into two groups
1) A normal group (controls) who is free of BC.
2) Cases with suspicious mammograms confirmed by biopsy.
h- Calculating breast cancer risk by using Gail model software
i- Calculating breast cancer risk by using Tyrer-Cuzick model software
j- Statistical analysis:
The main point was testing models’accuracy through calibration and discrimination.
Summary, Conclusion and Recommendations
80
The most important findings of our study
Calculating cancer risks in the study sample by models:
About 27% of the enrolled women had a Gail 5-year risk greater than average population risk and 24% had a lifetime risk higher than average population risk. Around 14% of enrolled women were high- risk of BC with a 5-year risk >1.66% which is the Gail cut off point based on FDA guidelines and previous literature. Regarding Tyrer-Cuzick, About 26, 27, 29% of the enrolled women had risks greater than average population risk in 5-year,10-year and lifetime scores respectively. Only 3.17% of enrolled women were high- risk of BC with a lifetime risk ≥ 20% which is the Tyrer-Cuzick cut off point based on American cancer society guidelines and previous literature. The 5-year, 10-year risks were noticed to be increased with age while lifetime risks decreased by increasing age.
Comparing cancer risks between BC cases and controls by models:
BC cases had significantly higher estimates than controls in respect to Gail 5-year, Tyrer 5-year and Tyrer 10-year risks. However, lifetime risks were not statistically different between both groups.
Testing models’ accuracy by calibration and discrimination:
o Three of these risk levels were well calibrated; the Gail 5-year, the Tyrer 5-year and 10-year risks meaning that the observed count of BC cases was nearly equal to the expected count. Gail 5-year, Tyrer 5-year and Tyrer 10-year risks showed a modest discrimination (AUC= 0.628, 0.628, 0.607 respectively). On the other hand, Gail and Tyrer lifetime risks failed to predict BC cases. This low performance may be mainly due to the included risk factors in these models which didn’t significantly affect presence of BC in our sample although their effect being well established in the literature. Moreover, some included factors which weighed much in these models were present in a low percentage in our population like extensive family history. Also, the most significant risk factors yielded from the bivariate and multivariate logistic regression analysis were not included in these models like age of first live birth.
o The standard cut off point for Gail 5-year risk (>1.66) was not the best and replaced by (> 0.6) with good sensitivity but low specificity meaning that a woman whose risk >0.6 would be considered a high risk. The model will be good negative. The best cut off point for Tyrer 5-year risk is > 0.9 with fair sensitivity and specificity (54, 67.31%). Tyrer 10 year risk had a cut-off point of (>1.8). By comparing best three models, there was no statistically significant difference of AUC between Gail and Tyrer 5-year risks. However, Gail may be better in terms of sensitivity with minimal missing of high risk women. Also, there was a statistically significant improvement in AUC for Tyrer 5-year over Tyrer 10-year risk. Therefore, short term prediction (5 years) is better than long term (10 years) which is better than lifetime prediction.
Testing models’ agreement:
There was a highly significant moderate level of agreement between all risk levels. Also, a significant positive high correlation was found among most of them.
Summary, Conclusion and Recommendations
81
6.2. Conclusion
from the present study, we concluded that:
To date, limited studies about comparisons and even testing existing breast cancer risk prediction models have been conducted. from our evaluation of the Gail and Tyrer-Cuzick models with their five risk levels, it appeared that they have a low accuracy in assessing BC risk in Egyptian women. Three of these risk levels namely Gail 5-year, tyrer 5-year and 10-year risks were well calibrated. Both models were well fitted to our data with no preference as our population was low risk including a low percentage of high risk groups as nulliparous, late maternal age at first live birth and positive BC family history which is included in details in Tyrer model raising its performance in high risk populations.
The aforementioned three risk levels had a significant but still poor discriminatory ability as mentioned by many studies. This could be explained as these models calculate woman’s risk according to established risk factors in the populations where models had been developed. But the situation is different here in our Egyptian population as models’risk factors were equally present in controls as in cases as women who sought screening were mostly those with positive family history. Additionally, novel risk factors should be included when calculating the risk especially psychological as anxiety, stress and depression as they were reported by a good percentage of cases. Since it was reported that nearly 60% of breast cancers can arise without any known risk factors (Madigan et al., 1995), depending only on known risk factors in calculating BC risk will often be misleading.
Though, existing models have limitations in their ability to accurately identify high risk women, they can still be used especially if no available mammography as included risk factors still play a role especially with increasing delayed childbirth in women, placing them at greater risk of breast cancer.