Electronic Health Records (EHRs) – Part 2:
Arguably, structured data can be as unreliable as unstructured data. Structured information can often be poorly translated into codable data, as supported by Eason and Waterson’s (2014) research results that showed frontline staff resorted to workarounds to accomplish their work goals when using EHRs. Likewise, unstructured data can be problematic; yet, it provides valuable insights. The difficulty lies in designing a process to gather the information required. Capturing the descriptive information to reflect patient-clinician communication and computable data for administration and research purposes requires new models.
There has already been an increased interest in text mining tools to analyse free text alongside structured data. St-Maurice, Kuo and Gooch’s (2013) experiment – combining natural language processing and primary care data to analyse a system – proved the possibility to assess both structured and unstructured data. However, it must be taken into account that both free text and structured data do not include everything.
Free text is a clinician’s interpretation of the patient, combined with their personal views and understanding. Structured data is limited to the information a patient provides. There may be information missed by both parties – information the clinician has not noticed, and information the patient has omitted or is even unaware would be critical. Another viable solution, explored by Anderson et al. (2015) is the use of EHRs to monitor more than just the basic information. Their research showed that, while few cases of suicidal ideation and attempt were documented in EHRs, increased documentation could improve their monitoring in the healthcare system.
Combining EHRs and free text can provide researchers with a better idea of the symptoms of the individual, how the individual has responded to treatments, and the context in which the individual’s mental illness emerged or become more prominent. By applying natural language processing (NLP) to EHRs, Perlis et al. (2012) concluded NLP application would enable accurate outcomes because it would process text into meaningful concepts based on a set of rules (Perlis et al., 2012). The more that can be drawn from this, the better it is for developing personalised medicine.
If well developed, results can be fed back in real time to alert the clinician treating the patient and use the information provided to build informative resources or software that can help both alter perceptions and find cures (Huang et al., 2014). For example, if a model was built, similar to that by Poulin et al. (2014) who were able to develop linguistics-driven prediction models to estimate the risk of suicide with 65% accuracy, this would have a substantial effect on the progress of providing a solution to some mental health problems.
Individually, all these databases have their advantages and disadvantages, and when linked together we are able to harness the power of big data even more. Nevertheless, there are issues with this:
Firstly, some of the data has not typically been collected for research purposes and, therefore, is vulnerable to problems. These problems span from a lack of clarity, resulting in misinterpretations, to omitted data which can affect future models.
Secondly, building a model capable of interpreting this data in one format is extremely difficult as free-text data is subjective.
Thirdly, while there is initial data (on the individual with a mental disorder), whether that be the prescriptions they have to take or when they were admitted into a hospital, no other information precedes this. Researchers are left stumped, with minimal information on how the individual is coping with their mental health and mental illness post clinical treatment or interventions.
Finally, since the diagnosis of people with mental disorders are not always documented because the individual chooses not to disclose any information to a GP, or the GP is unable to identify if the individual has a disorder, lots of data remain missing.
It is important to note that those in charge of the EHR databases and any research using the information from the database take into consideration the patient’s wishes and best interests. While full anonymisation of data may not be the possible (Fernandes et al., 2013), it is imperative the data is protected in other ways. For progress to be made, researchers, clinicians and patients must work together.
Written by Rodney
If interested in the references, or would like more information, contact us via email: firstname.lastname@example.org