Searching - What are the questions a researcher would like to ask?

[davidmiller]

Frist, let us reference the useful input from openhealthcare/opal#114

Last:

Please add questions that researchers may want to ask the search system so that we can then translate these into user stories/concrete implementation tasks.

[michaeledwardmarks]

Examples from openhealthcare/opal#114:

1 - Show me all patients who had community acquired pneumonia as an unchanged diagnosis 2 - show me all the patients where diagnosis was at any point PUO - and then show me what their final list of diagnoses were 3 - show me everyone where an HIV test was positive between 2014-2015 4 - show me everyone who was at somepoint tagged ID 5 - what percentage of blood cultures was an organism entered for 6 - show me the patients who went to nigeria and whose reason for travel was VFR 7 - what antibiotics did we give people with pyelonephritis

@maggiearmstrong is the queen of database searching on the current system. Suggest we ask her to outline some common requests she gets for searching the existing database - and would be helpful to know both requests that we can and can't currently achieve with the existing database so we have a benchmark.

[GabPoll]

I'll add to that another example. "What antibiotics did we treat a patient with an infection caused by organism X, resistant to antibiotic Y"? - a micro slanted question, but important in era of increasing antibiotic resistance. Essentially, a way to search for subfields within the "Investigations" modal.

Happy to take advice from @maggiearmstrong though.

[davidmiller]

Folks - these are super useful :) Do keep adding to the list if/when you think of 'em.

[michaeledwardmarks]

A few more questions.

These are less research questions more clinical audit/service delivery questions but important. Obviously some of the crunching would be done outside of the database but gives a sense of the kind of data we would want.

1) Of people where the clinical advice modal was completed (with any data): What was the reason for interaction What were the diagnoses What were the bugs Which of the auditable outcomes (checkboxes) were completed

2) Of people tagged ID-Liaison what was the total/average number of times clinical advice was documented. Over what time span Did Microbiologists also give advice (e.g also tagged to Micro and a microbiologist completed a clinical advice modal) Equally this question in reverse e.g of people tagged micro how often did ID-Liaison see/give advice

3) How many people had an indicator condition for HIV (e.g one of a list of diagnoses) Of these how many actually were offered an HIV test?

[michaeledwardmarks]

Some suggestions from Maddy. Again envisage that most of the number crunching would be done after extraction but gives a feel for what is of interest:

1- What is the variance in duration of antibiotic therapy and length of stay in specific diagnoses (say pyelonephritis or pneumonia) 2- What are the most common diagnoses in the ID liaison service 3- What is the age distribution of patients in each firm 4- What was the length of stay for IVDU patients

[michaeledwardmarks]

maddy made a suggestions so put here as a question related to this. Would we be able to automate some of the common audit type queries to run on a monthly basis - to effectively generate a monthly activity report e.g In the last month, for patients: Tagged ID-Inpatient Give the: Number of patients Age and sex distribution Diagnosis Length of stay

Alternative would be to have these searches defined and then for someone (? @maggiearmstrong ) to run the search each month.

Can imagine each team might have some key indicators that this would be helpful for as the basis for realtime audit.

[GabPoll]

A few other suggestions:

• of all the patients tagged "micro-haem" in last x months, how many of them are prescribed drugs "a" or "b"? (really targeted at caspofungin use, in case you're interested @michaeledwardmarks )
• of all the patients prescribed drug "a", what was their underlying diagnosis, or had a CT scan (with date extracted to compare to date of drug starting) (again looking at our predictor strategy for starting antifungals - no gold standard at the moment
• extract patients who had a set of investigations C idff antigen positive, toxin negative, and then subsequently C diff antigen positive and toxin positive. Would then want to pull out their a) diagnosis, b) PMHX, c) antimicrobials - again looking at risk factors for people who convert to toxin positive - once again evidence-light zone.

[davidmiller]

7 - what antibiotics did we give people with pyelonephritis

Data http://datapipes.okfnlabs.org/csv/html?url=https://gist.githubusercontent.com/davidmiller/9255218/raw/cdf1040780af85542a20a4ec7580d8828b9b29f9/antimicrobials.csv

[GabPoll]

Ha! Nice to see some data output. Obviously it's a very crude look, but it demonstrates a few important things:

• importance of duration of data by any user post interrogation
• importance of looking at dates where possible. Patients can receive several courses of antibiotics for different diagnoses
• again future possibilities of linking modals during data entry - e.g. When writing pyelonephritis, system prompts antimicrobial,used, or alternatively if write an antimicrobial, system asks for what diagnosis (offering options amongst those already entered, or a new one to enter). All conceptual, i know.

[GabPoll]

On my phone, so can't edit the entry. First point is curation, not duration!

[michaeledwardmarks]

Firstly - @davidmiller this is fantastic - really shows the power of the system.

Immediate thought is: 1) Answering this question demonstrates the need to cross-link data as @GabPoll mentions. To really answer this question you would need to have the list of diagnoses for each patient and then cross-reference with the antimicrobial prescribing. 2) Similar to @GabPoll other thought I wonder about an "Indication" field in the antimicrobials modal

I think we could use this dataset in the elCID review meeting on 13th March. I say this because

• Pyelonephritis is common and we will therefore have quite a bit of data
• We don't really know what we are doing!

If @davidmiller could generate (for same patient group: Diagnosis = Pyelonephritis) a) Diagnoses b) Investigations (specifically Blood Culture and Urine MC&S but can obviously give us everything)

That would let us test how the whole process of querying for a specific question, linking multiple sets of data and then interrogating that data works.

[davidmiller]

I have the zipfile containing all pseudonomised Pyelonephritis data as specified in #138 (Prior to considerations in #147 )

Can send you both copies if you want to play.

[davidmiller]

5 - what percentage of blood cultures was an organism entered for ?

76%

View Raw Data

Raw data with R script for generating plot/answer

[michaeledwardmarks]

Is the big column no entry? Amazing!

[michaeledwardmarks]

Can I get the full csv set for this too?!!!!! I'm in data ecstacy!

[davidmiller]

As this is fundamentally too conversational to be actionable, I move to close this. Am going to extract each use case as a user story ticket. Please do add more user focused extract tickets w. specific research q's. :)

[davidmiller]

2 - show me all the patients where diagnosis was at any point PUO - and then show me what their final list of diagnoses were

Not in this iteration. (Historical)

[davidmiller]

would be helpful to know both requests that we can and can't currently achieve with the existing database so we have a benchmark.

Did we ever figure out a list of these ? +10 would be super helpful.

[davidmiller]

1- What is the variance in duration of antibiotic therapy and length of stay in specific diagnoses (say pyelonephritis or pneumonia) 2- What are the most common diagnoses in the ID liaison service 3- What is the age distribution of patients in each firm 4- What was the length of stay for IVDU patients

IMHO these are all user-land answers, which we can already get if the dates weren't munged in the extract & the tags were historic.

[davidmiller]

of all the patients prescribed drug "a", what was their underlying diagnosis, or had a CT scan (with date extracted to compare to date of drug starting) (again looking at our predictor strategy for starting antifungals - no gold standard at the moment

@GabPoll Can you start a new issue with this? Also: I don't understand it :) can you explain what data/filter you'd need & what you'd want to do w. the data for me?