Health data constitute a significant source of medical information and can be used after suitable processing for diagnosis and treatment. They comprise a huge number of records, which may conceal significant patterns and dependencies. The vast amounts of data generated as well as the increased complexity of the health field make it hard to extract hidden information. A data warehouse is a proper solution for data integration because it permits the management and querying of data in various ways and forms. Retrieving data from multi-dimensional data warehouse supports the revelation of hidden and important information. Thus, applying data mining techniques to data warehouses that store healthcare data for knowledge discovery is appropriate because such data is already cleaned, grouped, and stored [1]. Thus, strategic decisions can be made to assist healthcare organizations.

Knowledge and experience are essential in doctors’ decision-making process. If decisions are made without proper caution, undesirable results can be encountered in the medical field, where the error tolerance is very low [2]. The use of data warehousing and data mining techniques for decision support has emerged as a new direction in the healthcare field. Decision support systems built using data warehousing and data mining techniques are information system applications that assist healthcare professionals in making the best decisions for patients by providing the most up-to-date information.

Cancer is a complicated disease that has many different types of variables and fluctuates rapidly. Additionally, there are many factors that should not be disregarded during diagnosis, follow-up, and treatment [1]. With the use of the decision support system based on data warehousing and data mining for cases of cancer, it will be possible to shorten the time needed for decision-making by healthcare providers. This will enable them to offer improved quality of care and achieve more positive outcomes.

According to the Global Cancer Observatory [3], a total of 18.1 million new cases of cancer occurred globally during 2018, and there were 9.5 million cancer-related deaths. The most frequently diagnosed cancers were lung (11.6%), breast (11.6%), and colorectal (10.2%), while cancer deaths were mostly due to lung (18.4%), colorectal (9.2%), and stomach (8.2%) cancers. If this rate continues to grow, it is estimated that 19.3 million of people will have cancer by 2025 globally.

Cancer-related research is extensive, expensive, and complex, and it involves various healthcare organizations. In cancer research, data analysis of cancer incidence is conducted by age, gender, region, ethnicity, as well as economic and social factors that contribute to the assessment of population health needs. The analysis of cancer data warehouses using data mining techniques may discover hidden relations among patients’ data, cancer treatment, and disease surveillance [4]. It is important to note that cancer care services are intrinsically multi-disciplinary, involving primary care physicians, pathologists, oncologists, and surgeons [5]. The efficacy of these services will be increasingly affected by the shortage of oncology professionals and the limited number of specialists in complex scientific disciplines that involve data science, cancer treatment, and surveillance [6].

In this study, we developed a clinical data warehouse for lung and ovarian cancer data provided by the United States National Cancer Institute (NCI; https://www.cancer.gov). These data include demographic information of the patient, past medical history, general health status, treatment modalities, and cancer characteristics [7]. The main contribution of this research is that for the first time the constellation schema is used to combine data from different cancer types to extract useful information related to hidden associations among medical features by data querying and analysis. The fact constellation schema consists of multiple fact tables, which share dimension tables. In fact, the fact constellation schema consists of more than one-star schema at a time by providing a flexible schema in which complex queries can be used to access data from the data warehouse. Other cancer data warehouse studies have been reported in the literature; however, they have concentrated on a specific type of cancer [8–13]. Our proposed system is designed to include data from both lung and ovarian cancer to find some possible hidden correlations among attributes. The aim of this study was to develop a clinical data warehouse for lung and ovarian cancer to be further used in data mining methods and decision support systems.

There are data warehousing, data mining, and decision support system implementations for healthcare information systems worldwide. Gorgionne et al. [10] discussed how data warehousing, data mining, and decision support systems can reduce the national cancer hardship related to pharyngeal cancers. Their proposed system evaluates the efficacy of specified treatments and interventions with the formulations, detects cancer patterns in general and special populations, organizes relevant claims data, and formulates models that explain the patterns.

Abidi and Abidi [14] introduced an integrated clinical evidence system designed to enhance clinical evidence with technology-mediated clinical evidence. Wu et al. [15] demonstrated that the integration of clinical decision support into computer-based patient records can enhance patient safety, reduce medical errors, improve patient outcomes, and decrease unwanted practice variations.

Bellaachia and Guven [9] used data mining techniques to predict the survivability rate of breast cancer patients. They implemented back-propagated neural network, naive Bayes and C4.5 decision tree algorithms. They found that the C4.5 algorithm’s performance was much better than that of the other two techniques. Wah and Sim [12] developed a clinical data warehouse for lymphoma to improve the quality of diagnosis and treatment recommendation decision-making.

Sheta and Eldeen [16] evaluated the architecture of a healthcare data warehouse specific to cancer diseases. They built the cancer data warehouse to integrate an operational database and medical files. Analysis of the data is facilitated by using OLAP (online analytical processing) cubes and viewing multilevel details of the data. According to the model developed in [11], the naive Bayes method is the most effective way to predict patients with lung cancer. Arous et al. [8] integrated two different operational health systems that include pancreatic cancer data. They explained the challenges they faced because the data types in the databases differed. Ramachandran et al. [17] developed a clinical data warehouse to identify potential cancer patients, which uses classification, clustering, and prediction data mining technologies. After the preprocessing step, the data is clustered using a K-means clustering algorithm. Their research can help detect a person’s predisposition to develop cancer before going for clinical and lab tests.

The remainder of this paper is organized as follows. Section II describes the methods applied related to the data warehouse domain and the research procedures used. Section III presents the results and describes the implementation of the proposed clinical data warehouse. Finally, Section IV provides our interpretation of the results, concluding remarks, and future research plan.

In this study, a clinical data warehouse for both lung and ovarian cancer data was developed for data mining and clinical decision support. The data required for the study were obtained from the NCI’s prostate, lung, colorectal, and ovarian cancer screening trial. Three major tasks had to be performed to prepare the data, namely, data acquisition, storage, and information delivery as depicted in Figure 1 [16].

Pre-processing of data is an important phase in the implementation of data warehouse applications. For data quality control, it is vital to pre-process the data to ensure that it is clean and meaningful. Lung and ovarian cancer data were cleaned to remove mistakes, repetitions, and stored in the data warehouse as a first stage. It was ensured that the data warehouse would conform to the subject-oriented, integrated, time-variant and non-volatile conditions of the data warehouse approach presented in [18].

Data warehouses are capable of integrating massive data from various sources for analysis and querying purposes. The healthcare field can benefit significantly from data warehousing for the study, prediction, and treatment of cancer, and in general, for the improvement of the quality of human lives.

In this study, a clinical data warehouse was designed for lung and ovarian cancers with data collected from the NCI. The main goal of this research was to analyze and store data regarding two different cancer types in one single data warehouse. We ensured that the data warehouse would conform to the subject-oriented, integrated, time-variant and nonvolatile conditions of the data warehouse approach presented by [18]. Data were integrated through preprocessing, transformation, and selection operations. In the design of the data warehouse, the fact constellation schema model, a multi-dimensional data modeling approach, was used. This model succeeded in responding to complex queries, and the analysis of data was facilitated by using OLAP cubes and viewing multi-level data details.

Indisputably, the fact constellation schema is the most challenging data warehouse design architecture. For the first time, real medical data were used combining two different cancer types using fact constellation modelling to extract semantically useful information to improve the decision-making process for cancer patients. The present approach also provides the ability to access data in the data warehouse using complex queries because dimensional attributes are shared by a number of fact tables, unlike the star and snowflake schemas. The fact constellation schema is capable of dealing with complex systems because the relationships between fact and dimensional tables are more easily understood. The medical field requires advanced data analytical processing by studying real critical medical data and this work addresses this need.

We plan to integrate prostate and colorectal cancer data into the already built clinical data warehouse and then to develop a decision support system for adapting more data analysis and mining tools. Hence, we can analyze cancer diseases, find correlations among attributes and cost of treatment for these diseases, death rates for specific types of cancer, and the impact of particular drugs on each disease. We expect that the final implementation of the decision support system will greatly assist cancer informatics research.