In this video, we're going to overview the main architecture of the MIMIC-III database and how it links information between the ICU units in the hospital records. Also, we're going to highlight that a major processing step to develop the database was to the identify information and remove any sensitive field. We need to particularly pay attention to dates that they have been shifted relatively to protect the privacy of the subjects. Understanding these processes is important later on when we would like to extract accurate information from the database. Finally, I'm going to highlight the key requirements to comply with the database usage. Electronic health records are complicated. This is for several reasons. This is a [inaudible] of the measurements and the type of information and caudate. Some of this information can be medical images, lab tests, natural language diagnosis from doctors, medications, and hospitalization events. Here we see some example of a patient timeline. This patient has been hospitalized. During hospitalization there is a number of tests he undergo, blood test, vital signs checked. It could be medical images and so on. We see here that she has visited his general practitioner. We also see later on routine hospitals checkups. A single patient data are spread over multiple electronic health record with diverse representation. Another important issue is the meaning of measurements. As simple temperature measure may vary depending on whether it is taking from the mouth or the armpit. Putting all this together, we see that electronic health records are irregularly sampled. Their nature is varied and dynamic. So how we can design the schema of a database to encode this information. This database should be accessed simultaneously from doctors and other health care providers frequently and in a unified way. Interoperability is a key requirement. This involve enhanced quality, efficiency, and effectiveness of the health care system. Information should be provided in the appropriate format whenever is needed. We should eliminate unnecessary duplications. Database selection and it's matching schema architecture usually influences that effective management of medical data flexibility, scalability, query performance, and interoperability. Non-proprietary standardized models are necessary to build electronic health record systems which comply the requirement of interoperability. MIMIC-III is a good example towards this direction. It is the only freely accessible critical care database of its kind. The dataset spans more than a decade, which detailed information about individual patient care. Analysis is unrestricted once a data use agreement is accepted. Enabling in this way, clinical research and education around the world. Databases such as MIMIC-III play a key role in accelerating research in machine learning models and end enabling reproducibility studies. Here we see the overall architecture of MIMIC-III database as it is presented in the relevant publications. MIMIC-III database links the identified information across five intensive units at the hospital of Medical Center in Boston with the hospital electronic health record databases and the Social Security Death Index. The intensive care unit include the medical intensive care unit, the surgical intensive care unit, the coronary care unit, the cardiac surgery recovery unit, and the neonatal intensive care unit. During ICU stay, there are several signals that are monitored and these are the vital signs, there are waveforms. We have alarms, but there are also fluids and medications as well as progression reports noted from the doctors. On the other hand, data recording from the hospital will include billing details and it includes also International Classification of Disease codes which relates to the pathology and the symptoms of the patient during admission. It will include demographics of the patient, and it will also include other nodes, with relation to medical images, discharge summaries, and so on. Building the MIMIC-III database involved a key step of the identification. This is a process which needs to comply with the Health Insurance Portability and Accountability Act. Therefore, all the fields related to patient data identification has been removed. This includes his patient name, telephone number, and addresses. In particular dates, we're shifted into the future by a random offset for each individual patient in a consistent manner. Preserving interval is important. Therefore, dates cannot be completely removed or randomly change. At the moment, dates have been shifted between years 2000 and 100 and 2000 and 200. It is also important clearing date shifting to preserve seasonality, but also time of day and day of the week. It was also taking action to protect individuals with red age characteristics. For this reason, dates of birth for patients aged over 89, were shifted to obscure their true age and comply with the Health Insurance Portability and Accountability Act. Finally, three text fields such as diagnostic reports and physician notes were carefully processed in order to remove any protected health information. We should reemphasize that the researchers that use MIMIC-III data, should complete a course in protecting human research participants according to Health Insurance Portability and Accountability Act requirements. Users should comply with appropriate data usage and security standards and shouldn't make efforts to identify individual patients. Summarizing, MIMIC-III links data across hospital, ICU unit, and Death Registry. We're going to see in great detail how to extract information from MIMIC-III database that includes lab examinations, medication, coding of International Classification Disease, and vital signals. MIMIC-III is the only database of its kind which is freely available without major restriction for its use. Data in MIMIC-III have been carefully processed in order to remove any information that can identify patients and dates are shifted relatively to protect patients' privacy.
