Pattern discovery, prediction and causal effect estimation in treatment discontinuation
With a trained, computerized discontinuation predictor machine learning component (MLC), predict, based on an input time series, a time when a subject will discontinue a course of medical treatment; with a trained, computerized pattern behavior extractor MLC, extract from said input time series the top k discriminatory sequences via discriminatory sub-sequence mining (said top k discriminatory sequences differentiate between first and second classes of interest to provide a hypothesis for downstream analysis of a cause of discontinuing said course of treatment). With a trained, causal effect estimator computerized MLC, determine a reason why said subject will discontinue said course of medical treatment, based on said top k discriminatory sequences and additional data; and with a computerized user interface, provide said time and said reason why to a responsible party to initiate an intervention.
1. A method comprising:
with a trained, computerized discontinuation predictor machine learning component, predicting, based on an input time series, a time when a subject will discontinue a course of medical treatment;
with a trained, computerized pattern behavior extractor machine learning component, extracting from said input time series a top k discriminatory sequences via discriminatory sub-sequence mining by comparing two or more subgroups and performing mining in search of patterns that appear unequally in the two or more subgroups, wherein said top k discriminatory sequences differentiate between first and second classes of interest to provide a hypothesis for downstream analysis of a cause of discontinuing said course of medical treatment;
with a trained, causal effect estimator computerized machine learning component, determining a reason why said subject will discontinue said course of medical treatment, based on said top k discriminatory sequences and additional data;
with a computerized user interface, providing said time when said subject will discontinue said course of medical treatment and said reason why said subject will discontinue said course of medical treatment to a responsible party to initiate an intervention; and
initiating said intervention to maintain said course of medical treatment.
2. The method of claim 1 , further comprising carrying out said intervention.
3. The method of claim 2 , wherein said intervention comprises counseling said subject and continuing to administer said course of medical treatment to said subject.
4. The method of claim 2 , wherein said intervention comprises counseling said subject, ceasing said course of medical treatment to said subject, and administering an alternative course of medical treatment to said subject which is not subject to said reason.
5. The method of claim 1 , wherein said computerized discontinuation predictor machine learning component comprises a recurrent neural network, and wherein said predicting, based on said input time series, when said subject will discontinue said course of medical treatment comprises applying said recurrent neural network.
6. The method of claim 5 , wherein said recurrent neural network comprises a long short-term memory, and wherein applying said recurrent neural network comprises applying said long short-term memory.
7. The method of claim 6 , wherein applying said long short-term memory comprises processing an input sequence of treatment episodes by recursively applying a transition function to an internal hidden state vector, including computing an activation of said hidden state vector at a given time step based on a current one of said treatment episodes and a hidden state for a time step previous to said given time step.
8. The method of claim 1 , wherein:
said input time series include a collection of sequences for continuing said course of medical treatment and a collection of sequences for discontinuing said course of medical treatment, and
said discriminatory sub-sequence mining comprises determining a support left value and a support right value based on said input collections of sequences, and
said top k discriminatory sequences are extracted based on said support left value and said support right value.
9. The method of claim 8 , further comprising pruning to eliminate from said top k discriminatory sequences those patterns that are obtained by augmenting an existing pattern, where said existing pattern is at least one of shorter and more general than a corresponding one of said augmented patterns, and has a higher confidence of predicting a class than said corresponding one of said augmented patterns.
10. The method of claim 8 , wherein providing said time and said reason with said computerized user interface comprises displaying a combination of:
a bar plot to show coverage of said collection of sequences for continuing said course of medical treatment and said collection of sequences for discontinuing said course of medical treatment; and
a shaded scatter plot to show said top k discriminatory sequences.
11. The method of claim 1 , wherein determining said reason why said subject will discontinue said course of medical treatment comprises determining a treatment effect by subtracting an outcome for an untreated group from an outcome for a treated group, and removing confounders by inverse probability weighting.
12. A computer program product comprising one or more computer readable storage media having stored thereon:
first program instructions executable by a computer system to cause the computer system to predict, based on an input time series, a time when a subject will discontinue a course of medical treatment;
second program instructions executable by the computer system to cause the computer system to extract from said input time series a top k discriminatory sequences via discriminatory sub-sequence mining by comparing two or more subgroups and performing mining in search of patterns that appear unequally in the two or more subgroups, wherein said top k discriminatory sequences differentiate between first and second classes of interest to provide a hypothesis for downstream analysis of a cause of discontinuing said course of medical treatment;
third program instructions executable by the computer system to cause the computer system to determine a reason why said subject will discontinue said course of medical treatment, based on said top k discriminatory sequences and additional data;
fourth program instructions executable by the computer system to cause the computer system to provide said time when said subject will discontinue said course of medical treatment and said reason why said subject will discontinue said course of medical treatment to a responsible party to initiate an intervention; and
fifth program instructions executable by the computer system to cause the computer system initiate said intervention to maintain said course of medical treatment.
13. A system comprising:
a memory;
a non-transitory computer readable medium comprising computer executable instructions; and
at least one processor, coupled to said memory and said non-transitory computer readable medium, and operative to execute said instructions to:
instantiate a trained, computerized discontinuation predictor machine learning component, a trained, computerized pattern behavior extractor machine learning component, a trained, causal effect estimator computerized machine learning component, and a computerized user interface;
with said trained, computerized discontinuation predictor machine learning component, predict, based on an input time series, a time when a subject will discontinue a course of medical treatment;
with said trained, computerized pattern behavior extractor machine learning component, extract from said input time series a top k discriminatory sequences via discriminatory sub-sequence mining by comparing two or more subgroups and performing mining in search of patterns that appear unequally in the two or more subgroups, wherein said top k discriminatory sequences differentiate between first and second classes of interest to provide a hypothesis for downstream analysis of a cause of discontinuing said course of medical treatment;
with said trained, causal effect estimator computerized machine learning component, determine a reason why said subject will discontinue said course of medical treatment, based on said top k discriminatory sequences and additional data;
with said computerized user interface, provide said time when said subject will discontinue said course of medical treatment and said reason why said subject will discontinue said course of medical treatment to a responsible party to initiate an intervention; and
initiating said intervention to maintain said course of medical treatment.
14. The system of claim 13 , wherein said computerized discontinuation predictor machine learning component comprises a recurrent neural network, applied to predict, based on said input time series, when said subject will discontinue said course of medical treatment.
15. The system of claim 14 , wherein said recurrent neural network comprises a long short-term memory.
16. The system of claim 15 , wherein said long short-term memory is applied by processing an input sequence of treatment episodes by recursively applying a transition function to an internal hidden state vector, including computing an activation of said hidden state vector at a given time step based on a current one of said treatment episodes and a hidden state for a time step previous to said given time step.
17. The system of claim 13 , wherein:
said input time series include a collection of sequences for continuing said course of medical treatment and a collection of sequences for discontinuing said course of medical treatment, and
said discriminatory sub-sequence mining comprises determining a support left value and a support right value based on said input collections of sequences, and
said top k discriminatory sequences are extracted based on said support left value and said support right value.
18. The system of claim 17 , wherein said trained, computerized pattern behavior extractor machine learning component is further operative to prune to eliminate from said top k discriminatory sequences those patterns that are obtained by augmenting an existing pattern, where said existing pattern is at least one of shorter and more general than a corresponding one of said augmented patterns, and has a higher confidence of predicting a class than said corresponding one of said augmented patterns.
19. The system of claim 17 , wherein said computerized user interface provides said time and said reason by displaying a combination of:
a bar plot to show coverage of said collection of sequences for continuing said course of medical treatment and said collection of sequences for discontinuing said course of medical treatment; and
a shaded graph plot to show said top k discriminatory sequences.
20. The system of claim 13 , wherein said trained, causal effect estimator computerized machine learning component is operative to determine said reason why said subject will discontinue said course of medical treatment by determining a treatment effect by subtracting an outcome for an untreated group from an outcome for a treated group, and removing confounders by inverse probability weighting.