From: Practicing precision medicine with intelligently integrative clinical and multi-omics data analysis
Number | Potential pitfalls |
---|---|
1 | Uneven distribution of informatics resources. |
2 | Integration of biomedical data located among heterogeneous sources. |
3 | Hazards in dehumanization of healthcare data. |
4 | Handling of extensively available irrelevant, error prone, and missing data. |
5 | Intelligent and user-friendly interface development. |
6 | Applying regulations and policies for data collection, usage and sharing. |
7 | Harmonizing big data with the definitions of clinical phenotypes and diagnosis. |
8 | Inflexible EHR database schemas not geared for precision medicine. |
9 | Lack of data availability on social determinants of health. |
10 | Unstandardized genomics tools and modifications in their versions and outcome format. |
11 | Overloaded Data generated during unnecessary follow-up diagnoses and treatments. |
12 | Augmented computational complexity with increasing number of attributes. |
13 | Slow SQL based high volume data processing speed. |
14 | Determining optimal parameters and understanding structures of AI and ML algorithms. |
15 | Handling continuous explanatory variables with more than two levels and understanding odds and probabilities in AI and ML algorithms. |
16 | Possibility of too many overfitting attributes in AI and ML algorithms. |
17 | Handling redundant attributes, distribution of statistically independent attributes, and management of class frequencies affecting accuracy. |
18 | Reduced evidence and reproducibility. |
19 | Correct predictor variables selection, and evidence-based observational data analysis and screening. |
20 | Gaining confidence of clinicians at AI produced results. |
21 | Ethical and social issues related to healthcare data collection, privacy and protection. |