-
Notifications
You must be signed in to change notification settings - Fork 0
Competency questions
Ethan Hartzell edited this page Jun 6, 2016
·
6 revisions
Validation of the DeepPhe models requires consideration of the types of questions that they might (and might not) be able to answer. Although such efforts are almost by definition limited in their scope and biased in their content, we attempt to list some questions that might be asked by cancer researchers. These questions might be based in our discussions with collaborators, review of the literature, or other sources.
- CLINICAL CRITERIA: Patients with certain diagnoses, demographics, or other clinical criteria
- EVENTREL: Identification of patients with specified relationships between sentinel events
- STRATIFICATION: Comparison of outcomes, treatments, etc. by cohort
- TRIANGULATION: Cross-source triangulation
- SCHEMA: Identification of available information
- Patients with certain diagnoses or other clinical criteria. Details of care may be needed for analysis, but temporal relationships between those details are not part of the filter.
- Possibly includes trends in lab values
- Often used as non-cancer controls
- Find demographic details, treatment details, reproductive information (age of first cycle, contraceptive use, number of pregnancies/births, number of abortions, etc), and autoimmune/inflammatory disease history for healthy patients or patients with endometriosis.
- Identify patients with increasing or decreasing levels of a biomarker (ca125).
- Find patients with events in specified temporal relations
- to each other (i.e., treatment before surgery)
- relative to some fixed or external time point
- after some time defined relative to patient's birth
- after some period of a change in practice or policy
- Possibilities include overlaps in the temporal extents of the events, and events separated by an event of a given bounded duration.
- In reviewing a database of breast cancer patients, find those patients who were given chemotherapy within 8 weeks of their death.
- What changes in genomic activity (i.e. tumor expression or mutation)might be associated with taking tamoxifen Progression-free or overall survival
- Recurrence
###Relevant Information Needs
- Patient demographics
- Age
- At time of observation
- At diagnosis
- Weight
- Height
- Smoking?
- Race
- SES
- Menopausal status
- Reproductive history
- Family History
- Medication History
- Previous surgeries
- Pre-existing conditions (hypertension, etc)
- Reasons for death.
- Time-stamped events, either instant or interval
- First Diagnosis
- Tumors
- Primary
- Secondary
- Recurrences - by types and location
- RECIST Criteria
- Surgical Margins?
- Size
- pre and post operative
- Histology type
- Number of positive lymph nodes
- Nottingham score (specific to breast)
- Pathological stages *Linkages between primary and secondaryQuestion
- Kinds of metastases
- Extent ( residual)
- Status (stable or not)
- Medications
- Starts, Stops
- Means of delivery
- Reasons for therapy
- Reasons for stopping
- Sensitivity
- Patient denies
- Reduced effect
- Number of cycles
- Chemotherapy
- Agent
- Metastatic
- Neoadjuvant
- Adherence to standard regimens
- Doses
- Radio therapy
- Bisphosphonates
- Reasons for timing/delays
- Side effects
- Major clinical events (i.e., death)
- Diagnoses - potentially hierarchical in varying degrees of specificity
- Type of cancer
- Age at time of diagnosis
- ICD-9/10 codes
- Imaging Diagnostics
- Dye-constrast wash-in/wash-out rate.
- BIRADS
- Hormonal changes
- Tests
- Specimen collection
- Genomic information, both normal and for tumor
- Genotypes
- H Score
- Her2
- Patient and tumor specific
- Expresion profiles
Given two sets of patients similar in key respects, compare certain outcomes based on stratification of categorical values such as care, phenotype, etc.
- Comparison of whole brain vs. stereotactic radiation in HER2 patients. Which survive longer? Counter-intuitively, whole brain. The resulting hypothesis is that the stereotactic radiation might have been creating micro-metastasses
- What portion of BRCA patients with PALB2 were given PARP inhibitor therapy?
- Does anything predict response to Eribulin?
- Number of skin lesions with lobular cancer as opposed to ductal?
- Why do patients switch treatments?
- What portion of BRCA patients with PALB2 got inhibitor therapy?
- Indication of factors that might be used to stratify a seemingly homogenous group.
- Treatment regimens
- Reasons for switching medications
- Intolerance/toxicity
- Progression - reasons for disease progression
- Failure to respond
- Recurrence/no benefit
- metastases
- acquired resistance
- Reasons for switching medications
- loss to follow-up
- Genotypes - expressed in terms of familiar clinical models (HER2, etc.)
- Medications
- Survival and tumor progression categorizations, based either on broadly accepted or arbitrary groupings.
- Comorbidities
- Adverse reactions
- Nursing notes
- Patient comments
Some information cannot be interpreted on the basis of any one source. Integration of related information from multiple sources is required to develop full understanding.
- Physician charts contain information regarding which medications are ordered, but nothing about whether or not those medications were actually taken.
- Which patients had medications that were ordered (as per physician charts), but not administered (as per Medication Administration Records (MARs) or nursing records)?
- Provenance of assertions with links back to original source for review and confirmation.
- Metadata on sources of comments
Before asking deeper questions, want to know what information is available on which patients. Given a sample of patients identified by restricting ranges of an input set of attributes a1, a2,.. ,and, how many of them have values from other attributes b1...bn?
- For which patients do I have a valid date of death?
- Schema of available data associate with patients of a given type - what are the range of data that might be available on any of these patients?
- This could be seen as a type of stratification/comparison query.