DA BIG DATA a PAZIENTE Carlo Combi. Department of Computer Science, University of Verona, Verona, Italy

Transcription

1 Data Warehouse and OLAP systems for Healthcare Ricadute dei Database e dello sharing dei dati in MG sulla ricerca statistica sanitaria: resoconto di esperienze DA BIG DATA a PAZIENTE 2015 Carlo Combi Department of Computer Science, University of Verona, Verona, Italy C.Combi (UniVR) DW and Healthcare 1 / 34

2 1 Introduction 2 Data Warehouses and OLAP 3 Data Warehouses and Clinical Domains 4 Big Data and Healthcare C.Combi (UniVR) DW and Healthcare 2 / 34

3 Introduction Introduction Definition Information = value-increasing asset, needed to effectively plan and control decision-based activities, as diagnosis, therapy planning, monitoring, health care management. Unfortunately data information. Having a huge amount of data makes it difficult to extract useful information. C.Combi (UniVR) DW and Healthcare 3 / 34

4 Data Warehouses and OLAP Data Warehousing Definition Decision Support System: set of techniques and software tools to extract information from a set of data stored in different sources. Among the Decision Support Systems, Data Warehouse Systems are those that are more established in the industrial world and could be suitably used also for biomedical data. Definition Data Warehousing: a collection of methods, technologies and tools to assist the knowledge worker (clinician, manager, nurse, epidemiologist, technician) to perform data analysis aimed at improving decision making and information assets. C.Combi (UniVR) DW and Healthcare 4 / 34

5 Data Warehouses and OLAP Complaints We have a huge amount of data but we can not access it! Why patients with the same therapy are showing significantly different results? We want to select, combine and manipulate patient data in every possible way! Show me only what is important! Everyone knows that some clinical data are not correct! C.Combi (UniVR) DW and Healthcare 5 / 34

6 Data Warehouses and OLAP ETL Tools The role of Extraction,Transformation and Loading tools is to feed a single data source, detailed, comprehensive, and of high quality, which may in turn feed the DW (Reconciliation). Stages of the reconciliation process: 1 extraction 2 cleaning 3 transformation 4 loading C.Combi (UniVR) DW and Healthcare 6 / 34

10 Data Warehouses and OLAP Multidimensional Model The model allows one to represent and query data stored in the Data Warehouse. A Data Warehouse is usually built incrementally and is composed of one or more data marts. A Data Mart may be composed of several Cubes. Facts of interest are represented in cubes, where: each cell of the cube contains numerical measures that quantify the fact; each axis of the cube represents a dimension of interest for the analysis; each dimension can be the root of a hierarchy of attributes used to aggregate data. C.Combi (UniVR) DW and Healthcare 10 / 34

11 Data Warehouses and OLAP Admissions Cube Figure: On 05/07/2009, 10 patients affected by ischemic heart disease were admitted to the cardiology department. C.Combi (UniVR) DW and Healthcare 11 / 34

12 Data Warehouses and OLAP Data analysis techniques Reporting Dashboard On Line Analytical Processing (OLAP) Data mining C.Combi (UniVR) DW and Healthcare 12 / 34

13 Data Warehouses and Clinical Domains Pharmacovigilance Activities C.Combi (UniVR) DW and Healthcare 13 / 34

14 Decision Support System for Pharmacovigilance MedDRA WHOART GIF RNF Med DBs ATC MSSO WHO GIF AIFA Various WHO Adverse Reactions Reports Drugs C.Combi (UniVR) DW and Healthcare 14 / 34

15 Pharmacovigilance: MedDRA Temporal Schema Code (T) Code (T) MedDRA term (LS) (1,N) Versioning (1,N) Version (LS) disjoint ISA System Organ Class Low Level Term High Level Term High Level Group Term ISA disjoint (1,N) (1,N) Comp pt hlt (VT) (1,N) Comp hlt hlgt (VT) (1,N) (1,N) Comp hlgt soc (VT) (1,N) (1,N) Non-Preferred (1,1)[1,N] Preferred (0,N) (1,1)[1,N] Comp pt soc (VT) Equivalence (VT) C.Combi (UniVR) DW and Healthcare 15 / 34

16 Pharmacovigilance: the Cube Conceptual Schema BroadTerm BroadTerm SMQ LowestLevelTerm LLTKey LLTName MedDRA PreferredTerm PTKey PTName NarrowTerm NarrowTerm SystemOrgan Class SOCKey SOCName Date Time Calendar EntryDate AdRStart AdREnd Report Report_ID PatientID Sex Age Patient Month MonthNumber MonthName Quarter Quarter Healthcare Structure HsKey StructureName Structure CommercialDrug CdKey CommercialName ATC Drug Drug DrugKey DrugName RootDrug Year Year Region RegionKey RegionName ATCKey ATC RootDrugKey RootDrugName C.Combi (UniVR) DW and Healthcare 16 / 34

17 Pharmacovigilance: OLAP Examples C.Combi (UniVR) DW and Healthcare 17 / 34

18 Pharmacovigilance: OLAP Examples C.Combi (UniVR) DW and Healthcare 18 / 34

19 Data Warehouses and Neonatal Metabolic Diseases The Regional Centre for Neonatal Metabolic Diseases (CRMMN) of Verona performs newborn screening for major hereditary metabolic and endocrine diseases galactosemia PKU (Phenylketonuria syndrome) Biotinidase deficiency deficiency of glucose-6-phosphate-dehydrogenase (G6PD) congenital hypothyroidism (IC) Leucine or maple syrup urine disease (MSUD) congenital adrenal hyperplasia (CAH) The screening is carried out on samples of blood taken from the heel of the newborn after 48 hours of birth C.Combi (UniVR) DW and Healthcare 19 / 34

20 Neonatal Metabolic Diseases: Qualitative assessment Fact schema The fact represented in this schema is the outcome of qualitative assessments, which records the result of the clinical validation of each quality examination carried out. The measures of interest are the number of times each outcome occurred of qualitative assessments (number of NO, VP, FN, DUB, etc.). C.Combi (UniVR) DW and Healthcare 20 / 34

21 Neonatal Metabolic Diseases: Doubt outcomes (DUB) grouped by weight In order to identify differences in weight groups, possibly highlighting different risk levels, the data mart represents the number of tests that indicate suspect values, according to the weight of the newborn. C.Combi (UniVR) DW and Healthcare 21 / 34

22 Neonatal Metabolic Diseases: Total outcome number We can see the total number of results by selecting only the dimensions date of birth, gestational age, type of examination, number of control and outcome. A total of 808,625 screening examinations has been performed. C.Combi (UniVR) DW and Healthcare 22 / 34

23 Neonatal Metabolic Diseases: Total DUB recall for patient born in 2010 Selecting only the DUB results and focusing only on patients born in 2010, it can be noticed that there were 1,495 recalled. C.Combi (UniVR) DW and Healthcare 23 / 34

24 Neonatal Metabolic Diseases: Drill-down on the type of examination By performing a drill down operation on the type of examination, it is possible to highlight how the quality tests carried out are grouped when the outcome is DUB. C.Combi (UniVR) DW and Healthcare 24 / 34

25 Neonatal Metabolic Diseases: Focusing on the congenital adrenal hyperplasia By selecting the exam type N170HPQUAL (qualitative exam for revealing the presence of CAH) and performing a drill-replace on the two members of the gestational age class level, we can see how the number of DUB for congenital adrenal hyperplasia (CAH) is related to the gestational age. C.Combi (UniVR) DW and Healthcare 25 / 34

26 Neonatal Metabolic Diseases: Focusing on the congenital adrenal hyperplasia The peak of DUB for CAH coincides with 36 weeks of gestation, but fell considerably with higher gestational ages. C.Combi (UniVR) DW and Healthcare 26 / 34

27 Neonatal Metabolic Diseases: Focusing on Leucinosis C.Combi (UniVR) DW and Healthcare 27 / 34

28 Data Warehouses and Drug Prescriptions We want to observe the pattern of three consecutive purchases of drugs, even the same product, for a patient; we consider a time span of 30 days as the maximum delay between the purchase of drug and the next one, and of 1 day as the minimum delay; The analysis focuses on the three most requested drugs for all patients in C.Combi (UniVR) DW and Healthcare 28 / 34

29 Drug Prescriptions: the Pattern Cube This cube represents patterns of subsequent prescriptions for a given patient, within thirty days between a purchase of a drug and the next one. C.Combi (UniVR) DW and Healthcare 29 / 34

30 Drug Prescriptions: Most requested drug triplets Viewing the 15 consecutive prescriptions at the second level ATC that verifies the most requested pattern, we note that the one that contains three times antacids, peptic ulcer and antimeteorics, exceeds any other triple with 20,384 requests. C.Combi (UniVR) DW and Healthcare 30 / 34

31 Drug Prescriptions: Most requested drug triplets Selecting antacids, peptic ulcer and antimeteorics and moving along the hierarchy to the fourth ATC level, we note that the most popular triplet consists of three identical descriptions of the acid pump inhibitors, with more than instances. C.Combi (UniVR) DW and Healthcare 31 / 34

32 Drug Prescriptions: Most requested drug triplets T1.1 (omeprazol) is the triplet most purchased, at the description ATC level. C.Combi (UniVR) DW and Healthcare 32 / 34

33 Big Data and Healthcare Big Data Technology and Healthcare Data Warehouses Multidimensional OLAP tools may rely on big-data storage and processing architectures; Methodologies are needed for tuning and integrating such tools within the healthcare domain. C.Combi (UniVR) DW and Healthcare 33 / 34

34 Big Data and Healthcare Big Data and Health: (Some) Issues and Research Lines Definition Big data is a term describing the storage and analysis of large and/or complex data sets using a series of techniques including, but not limited to: NoSQL, MapReduce and machine learning. Data anonymity and temporal clinical patterns; (Temporal) data integration: merging different data models; Temporal data processing and distributed architectures; Merging general data analytics tools and specific medical knowledge bases/models C.Combi (UniVR) DW and Healthcare 34 / 34