The interdisciplinary approach of the Complex Engineering Systems Institute (ISCI) has allowed us to address different public health problems with tools from Engineering, Artificial Intelligence, and Data Science, laying the foundations for the generation of public policies, supporting the practice of health personnel, and impacting the daily lives of patients. Here, we address three impact cases developed by Institute researchers:
- In Operations Management and Analytics, Susana Mondschein studies breast cancer prevention, focusing her research on reducing inequities. Based on its results, it generates relevant information bases to design public policies that impact the most vulnerable sectors of the population.
- Marcelo Olivares, leader of the Operations Management and Analytics group at ISCI, has developed a risk model for kidney graft failure for the Chilean population, which allows patients in the National Transplant System to be prioritized while reducing waiting times for an organ.
- Juan D. Velásquez belongs to the Data Science line and uses the potential of Artificial Intelligence to develop a tool capable of detecting probable cases of melanoma through photographs taken with a cell phone and determining which patients to refer for a biopsy.
Case 1: The importance of data to fight inequity in cancer epidemiology
Academic Susana Mondschein began her applied research work in cancer motivated to improve the prevention of this disease and its early detection. It initially focused on studying the epidemiology of breast cancer in Chile, which causes around 1,600 deaths each year, and is the leading cause of death in women of reproductive age (Source: International Agency for Research on Cancer, Globocan 2020).
Despite its incidence, to date there is no national cancer registry, so the researcher and her team set out to generate data that would allow them to draw conclusions and work on prevention.
The challenge was to build data on breast cancer in Chile between 2007 and 2018, based on an unprecedented methodology that made it possible to estimate the magnitude of the people who suffered from the disease.
To obtain this information, they accessed two anonymized databases provided by the Department of Statistics and Information (DEIS) of the Ministry of Health, with one containing diagnoses of hospital discharges and the second the mortality registry.
“One striking thing that we found is that in 2007, 3,785 cases of breast cancer were diagnosed, which contrasts with the 5,435 cases diagnosed in 2018. That is an increase of 43.6% over this period. When comparing the incidence by pension system, we found that the age-adjusted rate was 60.6 x 100,000 women in the case of Isapre (private healthcare companies) beneficiaries and 38.8 for Fonasa (the Chilean public healthcare system). This difference surprised us, as women who receive care via Fonasa have the right to complete treatment for free through the GES program (Explicit Guarantees in Health), a public policy that was created to level the playing field, and guarantee equity in rapid access to the best available treatment and with financial protection once the patient is suspected of having this disease, regardless of health insurance. However, something was suggesting that significant inequality continued to exist,”
says Susana Mondschein.
This inequality was also reflected when comparing the 5-year survival rate of patients who received treatment. In the case of women in the Isapre system it was 90.1%, while among Fonasa beneficiaries this percentage dropped 80.6%. The fatality figures for patients diagnosed with breast cancer are also consistent: 15.7% in Isapre versus 27.5% in Fonasa over the same period.
“We want to understand the scientific reasons behind this inequity, basing our conclusions on the data we have obtained, to generate the necessary public policies, and improve the situation. We have continued our research and confirmed that women in the Isapre system undergo more preventive mammograms than women affiliated with the public system, despite the fact that in certain age groups coverage of the exam is guaranteed. This data shows that resources must be put into increasing prevention and early detection in these segments of the population. Other factors that seem to play a role are differences in lifestyle, diet, and the use of hormone replacement therapy. From this basis, public policies that address obesity or a sedentary lifestyle can be evaluated, so that people are better prepared to fight the disease,”
concludes the researcher.
Case 2: Models to support the prioritization of kidney transplant patients in the National Transplant System
Around 300 kidney transplants are performed annually in Chile, but there is a waiting list of more than 2,000 people. Is there a way to optimize the process using the potential of data science?
This is the question that motivated researcher Marcelo Olivares and his team to design simulations to generate predictions regarding the compatibility and useful life of an organ available when transplanted in different patients, to increase the efficiency of the system and reduce waiting times for an organ.
“In our model, we identify the characteristics of the donor-recipient related to kidney transplant survival rates in the Chilean population. Given the substantial number of potential predictors relative to the sample size, we implemented an automated variable selection mechanism. An important point is that we work with a sample of 822 adult recipients, transplanted between 1998 and 2018. As far as we know, it is the largest kidney transplant database created in Chile to date,”
explains Marcelo Olivares.
To improve efficiency in allocating an available organ, seven variables were established that increase the risk of kidney transplant failure and a “weight” was assigned within the model. In the case of donors, age, male sex, history of hypertension, and history of diabetes. In the case of the recipient, the years they underwent dialysis and their history of organ transplant.
Based on the results of the simulation, it is possible to define priority cases within the National Transplant System.
Another way to support the reduction of the waiting list is by generating incentives to increase the list of living donors.
“The problem is that not any living donor can be allowed to give a kidney to another unknown person, since that could cause problems on a social level. Therefore, we are interested in implementing a model that has been successful in other countries, called the National Cross Transplant System. It seeks to improve the rate of living donors by allowing a centralized system to exchange organs between donors when they are incompatible with the patient associated with each donor. We believe that this is another way to continue improving our national transplant system from our research area,”
concludes the researcher.
Case 3: App based on Deep Learning algorithms detects early melanoma lesions through images
In Chile, approximately 800 people are diagnosed with melanomas each year and 300 die (source: International Agency for Research on Cancer, Globocan 2020). Studies show an increase in incidence, and this could be explained as a result of the deterioration of the ozone layer and the high levels of radiation that we have in the country.
A melanoma is the most aggressive of skin cancers, so early detection is key to the outcome of the disease. Normally, suspicious lesions are detected visually by dermatologists and a histopathological analysis is requested, through a biopsy. In this context, human error in the detection of this type of cancer is around 15%.
How can we support doctors who do not have dermatological training, those who treat in isolated areas, or those who detect a suspicious lesion in a patient but do not have the facilities to perform a biopsy?
This question brought together a team of experts in Artificial Intelligence, led by ISCI researcher and Professor of the Industrial Engineering Department of the Universidad de Chile, Dr. Juan D. Velásquez, and a team of health professionals led by Dr. María Flavia Guiñazu, neurologist and histopathologist from the Psychiatric Clinic, and dermatologist Fernando Valenzuela from the Department of Dermatology, both from the Clinical Hospital of the Universidad de Chile, to come up with an innovative solution.
“Since the detection of melanoma is visual in the first instance, we created a tool capable of identifying patterns of suspicious lesions on the skin through simple images, such as with a cell phone, determining on a scale the probability of a melanoma. In this way, the patient is only referred to biopsy the lesion if the probability is high. This tool has a high impact, as it supports the patient referral process and allows low-cost prioritization of which injuries to study, without the need to overload the system,” says the researcher.
How was this tool designed?
In the first instance, the team compiled a European database of images of more than 35,000 melanomas and with them trained a Deep Learning algorithm capable of emulating the neural networks of the human brain, to learn to recognize patterns between the photographs of the lesions.
However, they faced the challenge of adapting the algorithm to the national reality, since the phenotype of the Chilean population is influenced by the territory in which they live. With this objective, they calibrated the model specifically for Chile, adding more than 500 images of lesions extracted from the databases of the Clinical Hospital of the Universidad de Chile.
To validate the process, they considered a universe of 100,000 samples, some of which included melanoma, and fed 70% of the data into the tool, before taking 20,000 known images to calibrate the algorithm. They then tested the tool with 10% of the remaining images, obtaining an accuracy of over 85% in detecting melanoma by visual patterns, with high specificity and coverage.
“We are generating knowledge in Chile for Chileans and that is a paradigm shift. Health solutions cannot be easily imported from other countries, because they do not adapt to our own needs. For that reason, one of our proposed challenges is to generate a bank of skin lesions made up of 100% of samples with a Chilean phenotype, thus increasing the precision of the detection tool. It should be noted that our proposal does not replace doctors, but rather supports them. All images used for training the algorithm were evaluated and labeled by dermatologists around the world. Therefore, when uploading a local image, it is like a medical meeting with hundreds of dermatologists looking at the lesion and comparing it to those that have already been labeled as melanoma. This tool allows the doctor to perform better when making a diagnosis,”
concludes Juan D. Velásquez.
These three lines of research show that uniting engineering with expert knowledge of healthcare at the level of administration and medical specialists and working as a team with a systemic view makes it possible to generate base knowledge for the construction of public policies and technological innovations that impact medical practice and the quality of life of patients.