Utilizing HMO Claims Information and a Tree-Based Output Measurement for Medication Security Reconnaissance

Using hmo claims data and a tree based scan statistic for drug safety surveillance l.jpg
1 / 43
0
0
1360 days ago, 642 views
PowerPoint PPT Presentation
furthermore, Harvard Pilgrim Health Care. Upheld by gift HS10391 from the ... (AHRQ) to the HMO Research Network Center for Education and Research in Therapeutics ...

Presentation Transcript

Slide 1

Utilizing HMO Claims Data and a Tree-Based Scan Statistic for Drug Safety Surveillance Martin Kulldorff Department of Ambulatory Care and Prevention Harvard Medical School and Harvard Pilgrim Health Care

Slide 2

Supported by grant HS10391 from the Agency for Healthcare Research and Quality (AHRQ) to the HMO Research Network Center for Education and Research in Therapeutics (CERT) as a team with the FDA through Cooperative Agreement FD-U-002068 . Extend Collaborators: Richard Platt, Parker Pettus, Inna Dashevsky, Harvard Medical School and Harvard Pilgrim Health Care Robert Davis, CDC and so forth

Slide 3

Note of Caution Methodological Talk Substantive results demonstrated are exceptionally preparatory from the main early testing period of the venture.

Slide 4

Basic Idea Drug security observation is vital, since a few medications may bring about unsuspected antagonistic occasions (e.g. Thalidomide) Use HMO information on medication dispensings and conclusions of potential antagonistic occasions Data mining: For a specific finding, assess all medications For a specific medication, assess all judgments

Slide 5

HMO Research Network: Center for Education and Research in Therapeutics Fallon Community Health Plan (Massachusetts) Group Health Cooperative (Washington State) Harvard Pilgrim Health Care (Massachusetts, grantee association) Health Partners (Minnesota) Kaiser Permanente Colorado Kaiser Permanente Georgia Kaiser Permanente Northern California Kaiser Permanente Northwest (Oregon) Lovelace (New Mexico) United Health Care

Slide 6

HMO Data #HMOs: 10 Members: ~ 10.7 million Women: 51% Age <25: 34% Age 25-65: 53% Age 65+: 13% One year maintenance: ~80%

Slide 7

Three Major Methodological Issues Granularity: Is expanded hazard identified with a particular medication or a gathering of related medications? Modifying for Multiple Testing Calculating Expected Counts

Slide 8

Outline Tree Based Scan Statistic Application to Heart Attacks, Scanning All Drugs Calculating Expected Counts Future Plans

Slide 9

Nested Variables ecotrin Ì asprin Ì nonsteoridal calming drugs Ì pain relieving drugs intense lymphomblastic leukemia Ì intense leukemias Ì leukemia Ì malignancy

Slide 10

Drug Tree Based on American Society for Health-System Pharmacists (AHFS) Classification Level 1, with 18 bunches: Antihistamine Drugs (04) Anti-infective Agents (08) Antineoplastic Agents (10) Autonomic Drugs (12) Blood Formation and Coagulation (20) Cardiovascular Drugs (24) and so forth

Slide 11

Drug Tree Level 2: Anti-infective Agents (08) Amebicides (0804) Anthelmintics (0808) Antibacterials (0812) Antifungals (0814) Antimycobacterials (0816) and so on

Slide 12

Drug Tree Level 3: Anti-infective Agents (08) Antibacterials (0812) -Aminoglycosides (081202) -Antifungal Antibiotics (081204) -Cephalosporins (081206) -Miscellaneous Lactams (081207) etc

Slide 13

Drug Tree Level 5, bland medications (1009 aggregate): Anti-infective Agents (08) Antibacterials (0812) -Aminoglycosides (081202) -Gentamicin (081202-0002) -Geomycin (081202-0004) -Tobramycin (081202-0007)

Slide 14

A Small Two-Level Tree Variable Root Node Branches Leaf Drug A1 Drug A2 Drug A3 Drug B1 Drug B2

Slide 15

Granularity Problem Analysis Options Evaluate each of the 1009 non specific medication, utilizing a Bonferroni sort conformity for numerous testing. Utilize a higher gathering level, for example, level 3 with 184 medication bunches. Issue: We don't know whether a potential antagonistic occasion is expected to a littler or bigger medication aggregate.

Slide 16

Analysis Options The Other Extreme Take the 1009 non specific medications as a base, and assess each of the 2 1009 - 2 = 5.49 " 10 303 blends. Issue: Not all mixes are of intrigue.

Slide 17

Ideal Analytical Solution Use the Hierarchical Drug Tree Evaluate Different Cuts on that Tree

Slide 18

Cutting the Tree Cut Drug A1 Drug A2 Drug A3 Drug B1 Drug B2

Slide 19

Problem How would we manage the various testing?

Slide 20

Proposed Solution Tree-Based Scan Statistic

Slide 21

One-Dimensional Scan Statistic Studied by Naus (JASA, 1965)

Slide 22

Other Scan Statistics Spatial output measurements utilizing circles or squares. Space-time examine measurements utilizing barrels, for the early recognition of ailment flare-ups. Variable size window, utilizing most extreme probability instead of numbers. Connected for geological and transient malady observation, and in numerous different fields.

Slide 23

Tree-Based Scan Statistic H 0 : The likelihood of a conclusion after the administering of a medication is the same for all medications. H A : There is no less than one gathering of medications after which the likelihood of conclusion is higher . . . after different modification

Slide 24

Tree-Based Scan Statistic For every non specific medication we have: watched number of analyzed cases expected number of analyzed cases, balanced for age and sexual orientation

Slide 25

Tree-Based Scan Statistic 1. Examine the tree by considering every single conceivable cut on any branch. 2. For every cut, compute the probability. 3. Indicate the cut with the greatest probability as the no doubt cut (group). 4. Create 9999 Monte Carlo replications under H 0 , molding on the watched number of aggregate cases. 5. Look at the probably cut from the genuine information set with the no doubt cuts from the arbitrary information sets. 6. On the off chance that the rank of the no doubt cut from the genuine information set is R, then the p-esteem for that cut is R/(9999+1).

Slide 26

Log Likelihood Ratio c G = watched cases in the cut characterizing drug bunch G Ng = expected cases in the cut characterizing drug gather G C = add up to number of watched cases = add up to number of expected cases

Slide 27

Example: Acute Myocardial Infarction (AMI) Sample of Harvard Pilgrim Health Care Data 376,000 patients Years 1999-2003 2755 AMI analyze [Acute Myocardial Infarction = heart attack]

Slide 28

Results Most Likely Cut Drug(s): Nitrates and Nitrites (241208) Observed: 98 Expected: 7.3 O/E=13.4 LLR = 165.0, p=0.0001

Slide 29

Results Second Most Likely Cut Drug: Nitroglycerin (241208-0004) Observed: 77, Expected: 6.2, O/E=12.5 LLR = 124.3, p=0.0001

Slide 30

Results: Top 10 Cuts Obs Exp O/E LLR Drug(s) . 98 7.3 13.4 165.0 Nitrates and Nitrites (241208) 77 6.2 12.5 124.3 Nitroglycerin (241208-0004) 110 15.3 7.2 123.4 Vasodilating Agents (2412) 88 11.8 7.4 101.2 Adrenergic Blocking Agents (2424) 88 11.8 7.4 101.2 Adrenergic Blocking Agents (242400) 36 1.3 27.0 84.1 Clopidogrel (920000-0078) 209 74.6 2.8 83.6 Cardiovascular Drugs (24) 28 1.1 24.8 63.1 Isosorbide (241208-0003) 52 7.7 6.8 55.4 Atenolol (242400-0002) 32 2.9 10.9 47.5 Metoprolol (242400-0009) . p=0.0001, for all cuts

Slide 31

Results, Tree Format Obs Exp O/E LLR Drug(s) . 209 74.6 2.8 83.6 Cardiovascular Drugs (24) 110 15.3 7.2 123.4 Vasodilating Agents (2412) 98 7.3 13.4 165.0 Nitrates and Nitrites (241208) 28 1.1 24.8 63.1 Isosorbide (241208-0003) 0 0.0002 0 - Amyl (241208-0001) 77 6.2 12.5 124.3 Nitroglycerin (241208-0004) 5 6.7 0.7 - other 7 VA (2412xx) 88 11.8 7.4 101.2 Adrenergic Block Agents (2424) 88 11.8 7.4 101.2 Adrenergic Block Agents(242400) 52 7.7 6.8 55.4 Atenolol (242400-0002) 32 2.9 10.9 47.5 Metoprolol (242400-0009) 4 1.0 3.9 - other 11 ABA (242400-xxxx) 147 39.8 3.7 - other Cardiovascular Drugs (24xxxx)

Slide 32

Interpretation of Results People with cardiovascular issues are frequently taking cardiovascular medications and they are additionally at higher danger of AMI.

Slide 33

Observed and Expected Counts Exposed to sedate, had AMI Exposed to tranquilize, no AMI Unexposed to medicate, had AMI Unexposed to medicate, no AMI

Slide 34

Observed Counts Use just occurrence analyze Ignore the time after the episode analysis New medication clients versus common clients Length of medication presentation time window Cover holes in medication dispensings Use increase period before beginning to number

Slide 35

Multiple Drugs Individuals may all the while be "uncovered" to numerous medications Observed tallies are balanced for various medication utilize Expected checks are basically included for various medications, overlooking different medication utilize. Elective Assign every day as presented to at most one medication, selecting the most unprecedented one.

Slide 36

Comparison Group All non-uncovered days Remove days presented to cardiovascular medications while assessing cardiovascular analyses Censor people the day they begin utilizing a cardiovascular medication Other medication clients, evacuating non-tranquilize clients

Slide 37

Covariate Adjustments Age Gender HMO Temporal or regular patterns Frequency of medication utilize Disease chance elements (?)

Slide 38

Data Mining: A Cautious Approach Purpose is to create unsuspected signs Generated signals that must be deciphered from a clinical point of view. Signs might be unforeseen/vital or expected/insignificant. In the event that signs are not promptly expelled, they ought to be assessed utilizing standard epidemiological strategies.

Slide 39

Tree Scan Statistics: Future Developments Simultaneous utilization of various trees Scan analyze for a specific medication Simultaneous examining of medications and determinations utilizing two converging trees Drug-sedate association impacts Sequential checking of new medications Development of TreeScan programming

Slide 40

Final Remarks HMO information demonstrates guarantee for medication security reconnaissance The tree filter measurement can be utilized to take care of the issues of granularity and numerous testing Calculating watched and expected numbers is mind boggling and basic Data mining produces rather flags that should be affirmed/rejected utilizing different strategies Adopt other information digging techniques for HMO information

Slide 41

Reference Kulldorff M, Fang Z, Walsh SJ. A tree-based output measurement for database ailment observation. Biometrics, 59:323-331, 2003.

Slide 42

Comparison with Computer Assisted Regression Trees (CART) Four Simil

SPONSORS