My suggestions on coping with challenges in big health data challenges are

Missing Data: Since the missing data was not random, but from the intention of each study to omit or ignore meaningful data that might be useful for further analysis, there should be an agreement on what data should be collected in the similar topics of the study (eg. for cardiovascular research, there should be recommended variables that must be recorded) as a standard so the data (which might not be used to analyze in the current study) can be used in the future combined with other studies.

Data Analysis: It is true that while some clinicians or researchers have knowledge about traditional statistical tools, many still do not have adequate knowledge about handling and analyzing large datasets (even traditional statistical methods, many clinicians who are willing to do the research still do not understand them clearly). Newer techniques to handle and analyze big data, such as data mining and machine learning, should be trained in order to encourage clinicians and researchers to make use of big health data.

In terms of research and data collection, I think one of the ways to reduce the occurrence of missing data is to determine variables that are needed to help answer the research question. Another thing that needs to have thought about is the validation of each column. Oftentimes, the column is filled with wrong things that will be deleted in the data cleaning stages which lead to missing data. For example, putting “211” in the age section or having “Pregnant” in a “Male” patient.

In a questionnaire setting, we need to ask the questions clearly and avoid open-ended questions. Participants often skip the question if it is open-ended and/or if a long answer is needed.

As mentioned in the article, there are many distinctly different challenges.
The challenge of disease definition, data quality, and missing data. I suggest that we create a working group that gathers all cardiovascular disease specialized people to give the standardized disease definition, create only variables essential for further analysis, and create a user-friendly platform for EHR. After that, this working group should possess generalized knowledge to frontline workers (such as physicians, nurses, operators, or managers of the project), emphasizing the importance and benefit of those collected data to all members. The Working group should assign the project manager to verify the completeness of the data and feedback to the collector.
For unstructured data, we might have to reconsider its importance; if it is still essential, we have to structure it or make a score or tier it. For source availability challenges, we should start with the center or area willing to do so.
For data sharing, linkage, inconsistency, and security, we should create a database system that can provide or share the data with another researcher under the security circumstance. The database system should have the ability to collect data from a different source or input and match data if there were stored in various formats.
To interpret results, data analysis, and selection bias that might occur using the extensive data set, we should emphasize the correction of methodology and good clinical practice on the researcher to reduce the risk of bias if the data are collected very well.

Missing data in the study due to incomplete data in the EHR can be overcome by making EHR standards and standard variables mandatory and optional. The metadata standard is an agreement between practitioners and researchers. Data analysis problems can be overcome by involving statisticians so that in research, good collaboration between experts is needed.

I think we can reduce missing data by
1) Encouraging to use of terms in EHR standards and providing protocol/procedure of required data that need to be filled in the system. Training and make alignment within cross-functional teams and users is also crucial.
2) Set up the mandatory field and fixed answers, e.g. choices Female/male, if possible (for structured data)

For data analysis, it would be easier for statisticians to analyze data if they fully get necessary and clean data. However, I also agree with several opinions by classmates that good collaboration is important. Additionally, the team should consult an expert on analysis workflow/any possible errors and also training once initiating any innovative technology as suggested by Arwin.

– Missing Data : As mentioned in the article that “Since informations in EHR are non-systematically collected”. I think that to to cope with missing data, we might have improve method of data collection in both system and human. EHR should store as much structured data as possible, and minimize unstructured data collection. Applying terminology standard and reference information model such as SNOMED-CT or HL7 FHIR along with decent application design might help to improve the problem. Moreover, healthcare provider should record data in the same standard to lower missing data. That means organization should have policy and regulation about recording data.

– Selection bias and Data Analysis : I think that the biggest challenge in the problem is lacking of knowledge. We have to work as a multidisciplinary team, including data analyst specialist, to integrate variety of knowledge to solve the problem. Providing knowledge or training about data analyst for healthcare provider will be a good option too.

– Privacy and Ethical issue : Privacy and ethical are now considered in every aspect. Regulation, such as PDPA in Thailand and GDPR in Europe, must be followed to keep good privacy and good ethical acts.
–

As for the Privacy and Ethical Issue, while the use of “new social contract” for data utilisation seems to be realistic, as stated in the article, the establishment of a recently proposed model in which “authorised institutions” with data processing technology convert personal information into anonymized information using a defined technique, and provide the information to users as a business seems promising.

For the missing data issue, I think we need to understand all patterns of the data, identify characteristics of the data to determine the standard protocol and make an agreement with the team in terms of data collection.
In terms of data analysis and training, we need collaboration between researcher, statistician, or any other expertise, as others have said. Moreover, good clinical practice, PDPA, and privacy policy should be followed to develop the data security and conduct the research ethically.

As mentioned in the article there are different big health data challenges like missing Data, Privacy and Ethical Issues, and others. With the advancement of technologies in health sectors, there is a lot of progress going on and more to do to overcome the challenges.
Data are missing as clinicians don’t include them as they are of less importance and sometimes the patient doesn’t wish to disclose their information either. So to overcome this challenge, there should be a standard data protocol to include. Privacy, confidentiality should be protected in various ways like data encryption, multifactor security authentication, while dealing with data and review with the ethical board in dealing with ethical issues.

Missing data is the most common problem in every Electronic Health Records (EMRs). The data requirement in the EMR system may help in this situation, we need to fill the required data in order to submit them to the system. Some personnel who inputting the data into system may not aware of the need to submit the data. Hence, raising awareness of the personnel may be one of the choices to eradicate this issue. The statistical approach to missing data might also be a good choice too. For example, complete-case analysis, multiple imputation, etc.

The challenging issues in the big data are variety. One of these is the missing data which result from non-systematically selection both from collectors and informants. Therefore, the agreement on the data recieving should be performed before data collection process simultaneously with the designed data collection system which construct the required data. Another interesting issue in big data is data analytic and selection bias which researcher should be trained before the process as well as the participation/collaboration of the data specialist.

Taksin U.