Machine learning techniques for automatic evaluation of clinical trial data
Aspects of the subject matter described in this specification are embodied in systems and methods that utilize machine-learning techniques to evaluate clinical trial data using one or more learning models trained to identify anomalies representing adverse events associated with a clinical trial investigation. In some implementations, investigation data collected at a clinical trial site is obtained. A set of models corresponding to the clinical trial site is selected. Each model included in the set of models is trained to identify, based on historical investigation data collected at the clinical trial site, a distinct set of one or more indicators that indicate a compliance risk associated with the investigation data. A score for the clinical trial site is determined based on the investigation data relative to the historical investigation data. The score represents a likelihood that the investigation data is associated with at least one indicator representing the compliance risk.
1. A computer system implemented method for predicting a compliance risk of a clinical trial site, the method comprising:
obtaining, at the computer system, investigation data collected at the clinical trial site;
training, by one or more processors of the computer system and based on the investigation data collected at the clinical trial site, a set of machine learning models corresponding to the clinical trial site, comprising training a first machine learning model in the set of machine learning models to identify a first set of one or more indicators that indicate the compliance risk; and training a second machine learning model in the set of machine learning models to identify a second set of one or more indicators that indicate the compliance risk;
determining, by the one or more processors and for the first machine learning model, a first likelihood that the investigation data is associated with the first set of one or more indicators that indicate the compliance risk;
determining, by the one or more processors and for the second machine learning model, a second likelihood that the investigation data is associated with the second set of one or more indicators that indicate the compliance risk;
determining that the first likelihood has higher accuracy than the second likelihood;
based on determining that the first likelihood has higher accuracy than the second likelihood:
assigning a first weight to the first machine learning model; and
assigning a second weight to the second machine learning model, wherein the first weight exceeds the second weight; and
providing, by the one or more processors and for output on a user interface, an indication of the compliance risk of the clinical trial site based on the first and the second weights.
2. The method of claim 1 , wherein the compliance risk is associated with a subset of data records identified by the set of machine learning models as representing an adverse event specified by a regulatory agency associated with the investigation data.
3. The method of claim 2 , wherein the compliance risk indicates that all of the data records included in the subset of data records have not been reported to the regulatory agency.
4. The method of claim 2 , wherein the compliance risk indicates that one or more data records included in the subset of data records have not been reported to the regulatory agency.
5. The method of claim 2 , wherein:
the compliance risk indicates that the subset of data records are likely to be reported to the regulatory agency within a time period that exceeds a threshold time period for reporting the adverse event.
6. The method of claim 5 , wherein the threshold time period for reporting the adverse event is defined by (i) a first time point when the adverse event is discovered, and (ii) a second time point during when the adverse event is reported to the regulatory agency.
7. The method of claim 1 , further comprising:
combining the first likelihood and the second likelihood to determine the indication of the compliance risk of the clinical trial site.
8. The method of claim 7 ,
wherein a value of the first weight exceeds a value of the second weight; and
wherein combining the first likelihood and the second likelihood to determine the indication of the compliance risk of the clinical trial site comprises combining the first likelihood and the second likelihood based on the first weight assigned to the first machine learning model and the second weight assigned to the second machine learning model.
9. The method of claim 1 , further comprising:
determining that a combined likelihood satisfies a threshold value, wherein the combined likelihood is a likelihood combining the first likelihood and the second likelihood based on the first and the second weights assigned to a respective machine learning model; and
based on determining that the combined likelihood satisfies the threshold value, determining that the clinical trial site is a risk-associated clinical site.
10. The method of claim 1 , further comprises:
determining one or more attributes associated with the clinical trial site;
identifying, based on the attributes, one or more machine learning models in the set of machine learning models, wherein each machine learning model in the set of machine learning models is trained to identify, based on historical investigation data collected at the clinical trial site, one or more indicators that indicate the compliance risk of the clinical trial site.
11. A system comprising:
one or more computers and
one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
obtaining, at the one or more computers, investigation data collected at the clinical trial site;
training, by the one or more computers and based on the investigation data collected at the clinical trial site, a set of machine learning models corresponding to the clinical trial site,
comprising training a first machine learning model in the set of machine learning models to identify a first set of one or more indicators that indicate the compliance risk; and training a second machine learning model in the set of machine learning models to identify a second set of one or more indicators that indicate the compliance risk;
determining, by the one or more computers and for the first machine learning model, a first likelihood that the investigation data is associated with the first set of one or more indicators that indicate the compliance risk;
determining, by the one or more computers and for the second machine learning model, a second likelihood that the investigation data is associated with the second set of one or more indicators that indicate the compliance risk;
determining that the first likelihood has higher accuracy than the second likelihood;
based on determining that the first likelihood has higher accuracy than the second likelihood:
assigning a first weight to the first machine learning model; and
assigning a second weight to the second machine learning model, wherein the first weight exceeds the second weight; and
providing, by the one or more computers and for output on a user interface, an indication of the compliance risk of the clinical trial site based on the first and the second weights.
12. The system of claim 11 , wherein the compliance risk is associated with a subset of data records identified by the set of machine learning models as representing an adverse event specified by a regulatory agency associated with the investigation data.
13. The system of claim 11 , wherein the operations further comprise:
combining the first likelihood and the second likelihood to determine the indication of the compliance risk of the clinical trial site.
14. The system of claim 13 ,
wherein a value of the first weight exceeds a value of the second weight; and
wherein combining the first likelihood and the second likelihood to determine the indication of the compliance risk of the clinical trial site comprises combining the first likelihood and the second likelihood based on the first weight assigned to the first machine learning model and the second weight assigned to the second machine learning model.
15. The system of claim 11 , wherein the instructions, when executed, cause the one or more computers to:
determine that a combined likelihood satisfies a threshold value, wherein the combined likelihood is a likelihood combining the first likelihood and the second likelihood based on the first and the second weights assigned to a respective machine learning model; and
based on determining that the combined likelihood satisfies the threshold value, determine that the clinical trial site is a risk-associated clinical site.
16. The system of claim 11 , wherein the operations further comprise:
determining one or more attributes associated with the clinical trial site;
identifying, based on the attributes, one or more machine learning models in the set of machine learning models, wherein each machine learning model in the set of machine learning models is trained to identify, based on historical investigation data collected at the clinical trial site, one or more indicators that indicate the compliance risk of the clinical trial site.
17. A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, when executed, cause the one or more computers to perform operations comprising:
obtaining, at the one or more computers, investigation data collected at the clinical trial site;
training, by the one or more computers and based on the investigation data collected at the clinical trial site, a set of machine learning models corresponding to the clinical trial site,
comprising training a first machine learning model in the set of machine learning models to identify a first set of one or more indicators that indicate the compliance risk; and training a second machine learning model in the set of machine learning models to identify a second set of one or more indicators that indicate the compliance risk;
determining, by the one or more computers and for the first machine learning model, a first likelihood that the investigation data is associated with the first set of one or more indicators that indicate the compliance risk;
determining, by the one or more computers and for the second machine learning model, a second likelihood that the investigation data is associated with the second set of one or more indicators that indicate the compliance risk;
determining that the first likelihood has higher accuracy than the second likelihood;
based on determining that the first likelihood has higher accuracy than the second likelihood:
assigning a first weight to the first machine learning model; and
assigning a second weight to the second machine learning model, wherein the first weight exceeds the second weight; and
providing, by the one or more computers and for output on a user interface, an indication of the compliance risk of the clinical trial site based on the first and the second weights.
18. The non-transitory computer-readable medium of claim 17 , wherein the compliance risk is associated with a subset of data records identified by the set of machine learning models as representing an adverse event specified by a regulatory agency associated with the investigation data.
19. The non-transitory computer-readable medium of claim 17 , wherein the operations further comprise:
combining the first likelihood and the second likelihood to determine the indication of the compliance risk of the clinical trial site.
20. The non-transitory computer-readable medium of claim 19 ,
wherein a value of the first weight exceeds a value of the second weight; and
wherein combining the first likelihood and the second likelihood to determine the indication of the compliance risk of the clinical trial site comprises combining the first likelihood and the second likelihood based on the first weight assigned to the first machine learning model and the second weight assigned to the second machine learning model.
21. The non-transitory computer-readable medium of claim 17 , wherein the instructions, when executed, cause the one or more computers to:
determine that a combined likelihood satisfies a threshold value, wherein the combined likelihood is a likelihood combining the first likelihood and the second likelihood based on the first and the second weights assigned to a respective machine learning model; and
based on determining that the combined likelihood satisfies the threshold value, determine that the clinical trial site is a risk-associated clinical site.
22. The non-transitory computer-readable medium of claim 17 , wherein the operations further comprise:
determining one or more attributes associated with the clinical trial site;
identifying, based on the attributes, one or more machine learning models in the set of machine learning models, wherein each machine learning model in the set of machine learning models is trained to identify, based on historical investigation data collected at the clinical trial site, one or more indicators that indicate the compliance risk of the clinical trial site.