Meta-learning algorithm development to predict outcomes in patients with hepatitis C virus-related hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is the fourth most common cause of cancer-related death worldwide. Risk factors for HCC include chronic hepatitis B and hepatitis C, alcohol addiction, metabolic liver disease, and exposure to dietary toxins such as aflatoxins. More than 80 % of HCC cases occur in low-resource countries [1].

In Egypt, HCC is a serious health problem. It is the primary cause of death from cancer. Egypt has the world’s second-highest incidence of HCC, ascribed to the elevated prevalence and complications of the hepatitis C virus (HCV) [2]. HCC is the most prevalent malignancy in men, the 2nd most prevalent in women, and the most prevalent in both sexes combined [3]. This high incidence of HCC may be due to the increased frequency of HCV and its complications [4]. The incidence of developing HCC in HCV patients is 15–20 times greater than in uninfected patients. HCC rarely occurs without significant fibrosis or cirrhosis [5]. Advances in screening programs and diagnostic methods, together with the rising survival rate among cirrhotics, enable some patients to develop HCC [2].

Surveillance of patients with chronic HCV by ultrasonography enables the early detection of small HCC tumors and increases the chance of curative treatment. Early-stage HCC can be treated curatively by local ablation, surgical resection, or liver transplantation. Treatment selection depends on tumor characteristics, the severity of underlying liver dysfunction, age, other medical comorbidities, available medical resources, and local expertise. Catheter-based locoregional treatment is used in patients with intermediate-stage cancer [1].

Percutaneous ablation is recommended for small HCC tumors in patients with preserved liver functional reserve, according to the guidelines established by the American Association for the Study of Liver Disease and the European Association for the Study of the Liver [6]. Among the local ablation techniques, radiofrequency ablation (RFA) is minimally invasive and is the most commonly used modality [7], [8]. RFA results in a complete response comparable to liver resection [9], [10].

The prediction of prognosis and treatment outcomes in patients with HCC is complex. Expert clinicians struggle with high-dimensional criteria and the diversity of healthcare data in cases of HCC with underlying liver cirrhosis.

The computational power of machine learning (ML) can be used to explore the tremendous volumes of rich information found in electronic health records (EHRs) to discover hidden patterns and relationships in highly complex datasets. Decision trees are the favored technique for building easy predictive models, which are simple and fast with good accuracy, by allocating patients into subgroups [11].

A “soft” approach for reducing the need for computer power for data analysis is the introduction of meta-learning systems to select and rank the best-suited algorithms for different problems (datasets) [12]. These systems store historical experimental records (descriptions of datasets and algorithm performance) and, based on these records, evolve models to predict algorithm performance on new datasets. With these systems, an analyst does not have to evaluate a large number of algorithms with big data (only those with the best-predicted performance), saving computational and time resources. Thus, this biological data mining provides a launch pad for developing an applicable cloud computing platform that supports a new paradigm of data-intensive cloud-enabled predictive medicine [13].

In previous work, we implemented Data ML techniques to explore datasets to identify trends enabling the development of models for diagnosing HCC [14] and liver fibrosis [15], [16] and predicting therapeutic outcomes of patients with HCV utilizing simple laboratory data [17]. We also evaluated the performance of decision tree classifiers, assessing the correctly classified instances, recall, precision, and area under the curve [18].

The high prevalence of HCV-related HCC in Egypt motivated us to establish an HCC multidisciplinary clinic at Kasr Al-Aini Hospital, Cairo University, in 2009 as a referral center. This study aimed to summarize the experience of our HCC multidisciplinary clinic and investigate the predictive factors of treatment outcomes. We applied ML techniques to extract practical knowledge from large real-world datasets to provide a collaborative approach between clinicians and health informatics researchers. ML algorithms that integrate meta-learning frameworks to process massive amounts of complex HCC data and for ranking and selecting the best predictive algorithms can help scale up HCC management in a real-life clinical setting.

Our final goal was to develop a noninvasive algorithm to predict curative and palliative treatment outcomes for patients with HCC. This model needed to be reliable, based on routine clinical workups, and without imposing extra costs for additional examinations, especially in areas with limited resources, such as Egypt.

留言 (0)

沒有登入
gif