Background. According to the "Gartner eight sources of healthcare big data" [Gartner 2016], the challenging source named 'physicians free-text notes' shows up a remarkable 'Volume', consisting of 'terabytes of stored text', and a terrible 'Variety', said to be 'unstructured'. The amount of healthcare data now available and potentially useful to the care team reached 150 exabytes worldwide. Text analytics and cognitive computation is a research pathway that makes sense in hearing healthcare, to try to address the still unresolved needs for improving care pathways by the generic opportunities promised by big data analytics. In audiology, the 'physicians free-text notes' are human-generated data as the result of the most possible objective analysis of a number of roots-data coming from the patient-caregiver interactions. An audiogram, a medical history, an outpatient clinical notes, etc. All of these point out a highly varied audiological and frequently dispersed "roots-data" that are invisible to systems as they are unstructured. As a matter of fact, medical data coming as the result of a human synthesis do not benefit of the high degree of objectivity and/or standardization present in the data coming from machines only. The only one hope we have for automatizing - at least partially - such human synthesis is to go for analytics. This means also that every practicality, helping to improve - in significance as well as in robustness - the data where to start with analytics, is welcome. This contribution presents the first attempt in the hearing healthcare, to design and develop an easy-to-use pre-prototype clinical system for extracting, collating and analyzing audiological big-data from the diversified audiological free-text notes of the patient record. We will illustrate the results obtained from the evaluation phase of such a pre-prototype in the clinical management of aged people with hearing disabilities. Materials&Methods. Pilot evaluation of our system is done on a sample of patient records of elderly with CIs. Patient records included all the follow-up visits and comprised medical notes, audiometric test evaluations, questionnaires for self-assessment. Our system applies cognitive computing and big-data analytics, and leverages an ad-hoc lexicon developed for the hearing healthcare domain. Results: Our system extracts textual narrative information (e.g. from the etiology, audiological diagnosis, risk factors, surgical procedure to implant the hearing devices) and numerical information (e.g. audiometric tests, technical setup of CIs, questionnaires scores, etc.). Through cognitive computing and big-data text analytics, the system analyses medical notes written in plain language, extracts, by using textual and temporal contexts, the information that is used to plan the treatment pathway (such as audiological evaluation at CI follow-up visits, CI mapping, CI speech processor settings, etc.). Conclusions: The proposed pre-prototype clinical system uses text analytics for big-data to provide clinicians with a multi-source view of patient's hearing disability and it may be of some help to improve ongoing treatments of chronic conditions and proactive and preventive interventions. Grant: 'PNRCNR-AgingProgram_2012-18'.
An application towards text analytics and cognitive computing for clinical big-data in aged people with hearing disabilities
Gabriella Tognola;Alessia Paglialonga;
2017
Abstract
Background. According to the "Gartner eight sources of healthcare big data" [Gartner 2016], the challenging source named 'physicians free-text notes' shows up a remarkable 'Volume', consisting of 'terabytes of stored text', and a terrible 'Variety', said to be 'unstructured'. The amount of healthcare data now available and potentially useful to the care team reached 150 exabytes worldwide. Text analytics and cognitive computation is a research pathway that makes sense in hearing healthcare, to try to address the still unresolved needs for improving care pathways by the generic opportunities promised by big data analytics. In audiology, the 'physicians free-text notes' are human-generated data as the result of the most possible objective analysis of a number of roots-data coming from the patient-caregiver interactions. An audiogram, a medical history, an outpatient clinical notes, etc. All of these point out a highly varied audiological and frequently dispersed "roots-data" that are invisible to systems as they are unstructured. As a matter of fact, medical data coming as the result of a human synthesis do not benefit of the high degree of objectivity and/or standardization present in the data coming from machines only. The only one hope we have for automatizing - at least partially - such human synthesis is to go for analytics. This means also that every practicality, helping to improve - in significance as well as in robustness - the data where to start with analytics, is welcome. This contribution presents the first attempt in the hearing healthcare, to design and develop an easy-to-use pre-prototype clinical system for extracting, collating and analyzing audiological big-data from the diversified audiological free-text notes of the patient record. We will illustrate the results obtained from the evaluation phase of such a pre-prototype in the clinical management of aged people with hearing disabilities. Materials&Methods. Pilot evaluation of our system is done on a sample of patient records of elderly with CIs. Patient records included all the follow-up visits and comprised medical notes, audiometric test evaluations, questionnaires for self-assessment. Our system applies cognitive computing and big-data analytics, and leverages an ad-hoc lexicon developed for the hearing healthcare domain. Results: Our system extracts textual narrative information (e.g. from the etiology, audiological diagnosis, risk factors, surgical procedure to implant the hearing devices) and numerical information (e.g. audiometric tests, technical setup of CIs, questionnaires scores, etc.). Through cognitive computing and big-data text analytics, the system analyses medical notes written in plain language, extracts, by using textual and temporal contexts, the information that is used to plan the treatment pathway (such as audiological evaluation at CI follow-up visits, CI mapping, CI speech processor settings, etc.). Conclusions: The proposed pre-prototype clinical system uses text analytics for big-data to provide clinicians with a multi-source view of patient's hearing disability and it may be of some help to improve ongoing treatments of chronic conditions and proactive and preventive interventions. Grant: 'PNRCNR-AgingProgram_2012-18'.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
