OHDSI Home | Forums | Wiki | Github

OHDSI Study: Comparative effectiveness of alendronate and raloxifene in reducing the risk of hip fracture

Researchers
#1

We are pleased to announce the official start of the Save our Sisyphus Challenge – Hip Fracture study! See details on the wiki including study rationale, full protocol, and code.

Please provide any comments or suggestions over the next 1 week (by April 4).

If you would like to join as a contributing investigator, contributing to the analysis and write-up of this work, please let us know. So far we have participation from Hanyang University, UCLA, Columbia University, Stanford, Georgia Tech, IMS and Janssen. We would be delighted for you to join!

Thanks,

Yeesuk and Marc

1 Like
#2

Note that the because of a typo that needed fixing the new URL for the code is: https://github.com/OHDSI/StudyProtocolSandbox/tree/master/AlendronateVsRaloxifene

#3
#4

Marc the link to the code is not correct.

#5

Hi, @msuchard

There are some of combination drug of alendronate + cholecalciferol in Korea. Can this protocol distinguish pure alendronate and combination of alendroate + cholecalciferol? (If any alendronate drugs are picked by their ingredients and included in ‘excluded set’, alendronate alone or alendronate combination drug would not be distinguished. So alendronate group can be tend to have more cholcalciferol (I’m not sure my opinion is right).

Generally, alendronate drug has significant GI side effect. So compliance for alendronate is very low (as low as 20% in some reports…) So we can consider about this issue too.

#6

Hi @msuchard,

I am a data scientist from a data analysis company in China named “Palan DataRX”, I have lead several real world projects, and have some experience in comparative effectiveness. Currently, four of research work have been submitted to ISPOR and accepted. I would love to join this study and contribute to the statistical analysis work. :grinning:

Thank you!

#7

Hi @SCYou!

As you can see in the current definition of the target cohort, we select new-users based on the ingredient, so including combinations.

However, it is important to remember that drugs prescribed on the index date will be available for inclusion in the propensity model. If it is truly the case that alendronate and cholecalciferol go hand-in-hand in your data, I would expect cholecalciferol to be a strong predictor in the propensity model and any bias this might cause would be adjusted for by our propensity score adjustment. We actually have two good diagnostics to verify this happens correctly: the covariate balance plots (which will include the covariate for cholecalciferol) and the negative control distribution plot (which will show if residual bias exists after adjustement).

#8

Thanks for the response, @schuemie

I don’t worry about matching power of CohortMethod itself.

We need to add drug of alendronate into ‘excluded concept set’ for the variables matched. Then, aren’t combination drugs of alendronate + cholecaciferol excluded for the variables matched either?

#9

Good question!

On of the sources of covariates for the propensity model is the drug_era table. In this table, information about exposure is aggregated to the ingredient level. As you point out, information about alendronate will be discarded when building covariates, but all information about cholecaciferol will remain because the drug_era table does not contain information on combination products, just about the ingredients that may or may not be in combination products.

#10

Hi all, we have reformatted the protocol for our IPCI governance board and have submitted this.
The meeting of our board is next week (we did a last minute submission).

Hope we can contribute to the study.

Peter

#11

@schuemie
Yes, that’s what I wonder. Thanks for answering.

#12

Hi

I am from ZS associates and we will like to contribute on the statistical analysis part of the work. Do let us know what is required

Thanks
Manish

#13

Our claims data (recently converted to OMOP) spans only 365 days and the study requires at least 365+90 - so we cannot participate. :disappointed:

Before the study gets executed, perhaps this OHDSI study could be the first that in addition to table 1 (overview of population) also presents data on data quality.

A paper by Kahn at al argued for a table 1a that would present data for observational studies.
See it here.

The current code in terms of metadata only gets

createMetaData <- function(connectionDetails, cdmDatabaseSchema, exportFolder) {
    conn <- DatabaseConnector::connect(connectionDetails)
    sql <- "SELECT * FROM @cdm_database_schema.cdm_source"

Perhaps this could be extended (before final execution). (and I would be happy to contribute to coding that, since I am suggesting it)

The OHDSI [DataQuality]
(StudyProtocolSandbox/DataQuality at master · OHDSI/StudyProtocolSandbox · GitHub) package tries to reduce the number of attributes about a dataset (instead of all Achilles results data it has a much reduced subset of parameters) that a site could share and that is vastly less “sensitive”.
This could also be be just a tiny handful of parameters: (1) size of dataset population (as a size category [not exact size]) (perhaps just patients with at least 365+90 days in at least one obs period) (2) “level of inpatient-ness (% of outpatient visits out of all visits recorded); (3) level of claims-ness vs EHR-ness(e.g., % of patients with weight recorded) and some temporal span measures. I recently made a proposal to Data Quality Collaborative for MIAD (minimum information about a dataset) (as a similar concept to MIAME (minimum information about a microarray experiment).

#14

Feedback from Yiting Wang, a colleague at Janssen who has previous publications on the effects of bisphosphonates:

The protocol is very concise, still better to call out a couple specifics such as
Comparative effectiveness study in postmenopausal women (reflected in the cohort creation algorithm but not updated in the protocol, e.g., “gender” on page 6 of the 26-page PDF initially made me wonder if men were included)

“The target cohort and comparator cohorts will be stratified into 5 quantiles” (quintiles?) of the propensity score distribution (after trimming off the top and bottom 10% from the preference score distribution?)

• The extra 4-year age band of 46-49 (assuming standard 5-y age bands of 50-54, 55-59, …) likely contribute negligible number of hip fracture cases, while raising doubts on post-menopausal status, and whether hip fracture was osteoporotic or high-impact. Why not use age 50 and above? After all, osteoporotic hip fracture cases typically are ~70+ years old, and menopause typically occurs at age ~48-55.

• What is the rationale for requiring index date before 2012-02-01? How far before 2012-02-01? This does not seem to be related with proton pump inhibitors going OTC in the US, or osteonecrosis of the jaw being given an ICD-9 code.

• “Patients were excluded from consideration is (if) they qualified for both the target cohort and comparator cohort at any time in their record”. Does this exclusion criterion apply to post-index period, i.e., decision in the past based on info from the future?

• Hip fracture outcome restricted to inpatient diagnosis as either primary or sensitivity analysis?

• Non-hip, non-vert fractures that are more likely related to osteoporosis rather than accident for example, commonly include the following sites: rib, clavicle, humerus, radius/ulna, wrist, pelvis, or tibia/fibula. Fractures in fingers and toes may be included in sensitivity analysis.

• Vertebral compression fracture (VCF) outcome has to be interpreted with caution due to misclassification. E.g., Curtis (https://www.ncbi.nlm.nih.gov/pubmed/?term=19106733 ) found that more than half of incident VCFs were misclassified based a VCF diagnosis on any claim.

#15

Hi Marc,

This is Yonghui from UTHealth. We would like to join this research. We have the Cerner HealthFacts data and patient claim data from Blue Cross.

Thanks,
Yonghui

#16

Hi Vojtech,

Data quality remains very important. Could you make a git branch with a patch? At minimum, we could collect this information at a couple of sites.

best, M

1 Like
Standard table 1
#17

Hi Manish,

Great to have you involved. We’ll have R objects back from data partners shortly. Are there some specific analyses you’d like to undertake? Maybe you have some ideas about better meta-analyses across databases? You can take a look at the CohortMethod vignette to get an idea of the artifacts that each data partner returns.

best, Marc

#18

Study was apparently moved from sandbox folder to a Protocols “folder”.

Link that works is: https://github.com/OHDSI/StudyProtocols/tree/master/AlendronateVsRaloxifene

#19

@msuchard. I know Christian is a participant, but I would like to get our team involved. Can we join by running the study on our data?

#20

@mvanzandt:

I bring you in offline.

t