Trend analysis group updates
Good morning,
Happy Friday! I just wanted to follow-up with everyone regarding some things. Hannah, Bretta, and I spoke with the leadership group this morning to ensure that we were on the right track.
Here is the list of things I need from jurisdictions (associated attachments are listed below):
1. Validate your program against the ICD-10-CM validation dataset. Send results to Tom Largo (largot@michigan.gov)<mailto:largot@michigan.gov)>.
* PfS dummy data ICD10CM.xlsx
* Template for reporting dummy data results - trend group - case finding per Dx1 only.xlsx
* Coding Scheme for Validation Dataset for PfS Drug Hospitalizations.docx
* ICD-10-CM Provisional Transition Guidance.pdf
* CDC's Opioid Overdose Indicator Toolkit.pdf
2. Send me your quarterly count data as soon as possible. I'm still hoping for data from Louisiana, New Jersey, Tennessee, Washington, and Utah. The leadership group is expecting we have preliminary data to show at the Dec. 7 meeting, so please send me this as soon as possible.
* Overdose hospitalization 2010-2016-UPDATED-101617.sas
* Overdose_Hospitalization_with_R_updated.pdf
* QuarterlyHospitalizationCounts2010.2016_TrendAnalysisSubgroup_v3.xlsx
3. Begin using trend analysis program Hannah created. Send any comments or feedback as soon as possible. We will be working with data visualization group to determine how to best display results.
* Overdose Hospitalization 2010-2016-Trend Analysis.sas
Please let me know if you have any questions. Thank you!
Best,
Lia Pizzicato
From: Lia Pizzicato
Sent: Monday, November 6, 2017 5:38 PM
To: hannah.yang@mt.gov<mailto:hannah.yang@mt.gov>; gang.liu@health.ny.gov<mailto:gang.liu@health.ny.gov>; mmaurer@coconino.az.gov<mailto:mmaurer@coconino.az.gov>; sarah.nechuta@tn.gov<mailto:sarah.nechuta@tn.gov>; bretta.jacquemin@doh.nj.gov<mailto:bretta.jacquemin@doh.nj.gov>; melissa.jordan@flhealth.gov<mailto:melissa.jordan@flhealth.gov>; ebrutsch@utah.gov<mailto:ebrutsch@utah.gov>; Lily Tyndall Snow (Lily.TyndallSnow@LA.Gov<mailto:Lily.TyndallSnow@LA.Gov>); mamadou.ndiaye@doh.wa.gov<mailto:mamadou.ndiaye@doh.wa.gov>
Cc: Hal Johnson (hal@hjc-epi.com<mailto:hal@hjc-epi.com>) (hal@hjc-epi.com<mailto:hal@hjc-epi.com>); amypatel@cste.org<mailto:amypatel@cste.org>; Thomas Largo (largot@michigan.gov<mailto:largot@michigan.gov>) (largot@michigan.gov<mailto:largot@michigan.gov>); Pasalic, Emilia Sigrid (CDC/ONDIEH/NCIPC) (kse0@cdc.gov<mailto:kse0@cdc.gov>)
Subject: RE: Trend analysis group updates
Hi everyone,
As promised, here is the program for the trend analysis portion. Below is a description of the various parts of the program from Hannah, and I attached a few figures from Hannah for your reference. I apologize for the long email.
A few notes:
* Bretta ran the program using SAS 9.2, and some of the commands did not work in the older version. Please let me know if this would also be an issue for your jurisdiction as we need to think about whether we need a second program. Bretta is having some trouble with her SAS, but she can send out an alternative code once she resolves the issue for anyone using SAS 9.2. Or, if anyone else would like to volunteer to write a code for an older version of SAS, let me know!!
* Hannah did not write programs to look at the shift by age group and intent statistically as some of the categories are very sparse especially for the heroin poisoning indicator in her jurisdiction (see note #3 in her description). We will have graphs to show the data by age group and intent based on the data provided by jurisdictions for the count portion of the project, but do we as a group want to look at the shifts by age group and intent statistically? Or, do we think looking at the overall indicators (i.e. all drug poisonings, opioid excluding heroin, or heroin) statistically is sufficient?
* We are going to shift the deadline for the results of this analysis to December 15. This will hopefully give us time to address this program and give the data visualization group time to prepare spreadsheets. We should hopefully have spreadsheets out to everyone to fill in their data by November 30.
I'm thinking we might be due for a subgroup call as this analysis is getting a bit more complicated. We can see how the call on Thursday goes and then plan accordingly.
Thanks!
Best,
Lia Pizzicato
Description from Hannah:
1. Forecasting:
* OBJECTIVE: Predict the hypothetical scenario under which the transition had NOT taken place (the "counterfactual" scenario) and compare with the observed trend to evaluate the impact of the transition on drug poisoning counts.
* APPROACH:
* Use combined (not stratified) data set - "hospall_ts"
* Using only the ICD9-CM Era data points, predict the trend line (with 95% CI) forward for 5 quarters (Q4 2015 and Q1-Q4 2016) --> do this for all 3 indicators: all drug poisonings (H1), Non-heroin opioid (H2), and Heroin Poisonings (H3)
* To estimate the observed trends I used PROC AUTOREG
* Montana has a lot of Zeroes in our data, especially for indicator H3. I used PROC GENMOD with Poisson distribution
* See the attached graphs (H1, H2, and H3)
1. Interrupted time series analysis (segmented regression):
· OBJECTIVE: Quantify changes (level and/or slope changes) in Drug Poisoning counts following the ICD-10-CM transition & determine whether they are statistically significant
· APPROACH:
* Look at the combined (not stratified) dataset
· Proposed model:
Yt=β0 +β1(preslope)+β2(intervention) + β3(postslope)
· Yt = expected # of drug poisonings at time=t (H1, H2, and H3)
· Preslope- a continuous variable indicating time from start to beginning of transition --> models the pre-transition trend
· Intervention- a dummy variable coded 0 during icd9-cm and 1 during icd-10-cm (Intervention=1 starting from Q4 2015) --> models the level change attributable to the transition between the time points immediately before and after it
· Postslope- a continuous variable coded 0 up to the last time point in icd-9-cm (Q3 2015) and sequentially from 1 thereafter --> models the post transition trend
· β3- β1 = the change in trend (slope) before vs after (We wouldn't necessarily expect a change in the actual incidence of drug poisoning but maybe due to the differences in ICD 10-cm we may see a slope change??....)
* For montana for H1 I saw a significant level change immediately following the transition:
* There was also a slope change post transition but not statistically significant
Parameter Estimates
Variable
DF
Estimate
Standard
Error
t Value
Approx
Pr > |t|
β0
Intercept
1
298.0474
10.0962
29.52
<.0001
β1
preslope
1
-0.6996
0.7363
-0.95
0.3515
β2
intervention
1
-95.8565
26.3257
-3.64
0.0013
β3
postslope
1
3.1000
7.4075
0.42
0.6793
[cid:image003.jpg@01D34F05.CA159840]
1. Examine data by age group and intent
* OBJECTIVE: determine whether patterns seen in the combined dataset are seen in age stratified or intent stratified data
* I am not sure the best way to look at this. The Age stratified data for montana is VERY sparse especially for indicator H3.
* Here is just a simple graph of Montana data for indicator H1 by Intent, but I'm not sure the direction we want to go with it in terms of analysis
[cid:image004.jpg@01D34F05.CA159840]
From: Lia Pizzicato
Sent: Wednesday, November 1, 2017 9:43 AM
To: hannah.yang@mt.gov<mailto:hannah.yang@mt.gov>; gang.liu@health.ny.gov<mailto:gang.liu@health.ny.gov>; mmaurer@coconino.az.gov<mailto:mmaurer@coconino.az.gov>; sarah.nechuta@tn.gov<mailto:sarah.nechuta@tn.gov>; bretta.jacquemin@doh.nj.gov<mailto:bretta.jacquemin@doh.nj.gov>; melissa.jordan@flhealth.gov<mailto:melissa.jordan@flhealth.gov>; ebrutsch@utah.gov<mailto:ebrutsch@utah.gov>; Lily Tyndall Snow (Lily.TyndallSnow@LA.Gov<mailto:Lily.TyndallSnow@LA.Gov>); mamadou.ndiaye@doh.wa.gov<mailto:mamadou.ndiaye@doh.wa.gov>
Cc: Hal Johnson (hal@hjc-epi.com<mailto:hal@hjc-epi.com>) (hal@hjc-epi.com<mailto:hal@hjc-epi.com>); amypatel@cste.org<mailto:amypatel@cste.org>; Thomas Largo (largot@michigan.gov<mailto:largot@michigan.gov>) (largot@michigan.gov<mailto:largot@michigan.gov>); Pasalic, Emilia Sigrid (CDC/ONDIEH/NCIPC) (kse0@cdc.gov<mailto:kse0@cdc.gov>)
Subject: Trend analysis group updates
Hi everyone,
Happy November! I just wanted to follow-up with a few updates.
Quarterly Hospitalization Counts:
I've attached a new spreadsheet that includes a field to put the number of cases where intent is missing. I'm thinking this should only be an issue with ICD-9-CM, but please fill in the new spreadsheet if possible. Please send Quarterly Hospitalization Counts results by the end of the day on Friday, November 3. If you need more time, please let me know.
Validation Dataset:
I received the completed validation dataset for ICD-10-CM. Attached you will find:
* The validation dataset for ICD-10-CM
* Coding scheme for the dataset
* An excel spreadsheet to enter your results
* CDC's Opioid Overdose Indicator Support Toolkit - see specifically page 7
* Prevention for States guidance on generating morbidity indicators under ICD-10-CM
Please send completed spreadsheets using the validation dataset to Tom Largo (largot@michigan.gov<mailto:largot@michigan.gov>). He will check your answers. The validation dataset group should have a dataset for ICD-9-CM ready within the next week or so.
Trend Analysis Program:
Finally, Hannah and Bretta are finalizing the program for the trend portion of the analysis. It should be available by next week. The Data Visualization group will then use the completed program to create an excel spreadsheet for the results. We will send that out when it is ready.
Thank you and please let me know if you have any questions!
Best,
Lia N. Pizzicato, MPH
SAMHSA/CSTE Applied Epidemiology Fellow
Health Commissioner's Office | Opioid Surveillance, Epidemiology, and Prevention Program
Philadelphia Dept. of Public Health
Community Behavioral Health | Performance Evaluation, Analytics, & Research
Philadelphia Dept. of Behavioral Health and Intellectual disAbility Services
Happy Friday! I just wanted to follow-up with everyone regarding some things. Hannah, Bretta, and I spoke with the leadership group this morning to ensure that we were on the right track.
Here is the list of things I need from jurisdictions (associated attachments are listed below):
1. Validate your program against the ICD-10-CM validation dataset. Send results to Tom Largo (largot@michigan.gov)<mailto:largot@michigan.gov)>.
* PfS dummy data ICD10CM.xlsx
* Template for reporting dummy data results - trend group - case finding per Dx1 only.xlsx
* Coding Scheme for Validation Dataset for PfS Drug Hospitalizations.docx
* ICD-10-CM Provisional Transition Guidance.pdf
* CDC's Opioid Overdose Indicator Toolkit.pdf
2. Send me your quarterly count data as soon as possible. I'm still hoping for data from Louisiana, New Jersey, Tennessee, Washington, and Utah. The leadership group is expecting we have preliminary data to show at the Dec. 7 meeting, so please send me this as soon as possible.
* Overdose hospitalization 2010-2016-UPDATED-101617.sas
* Overdose_Hospitalization_with_R_updated.pdf
* QuarterlyHospitalizationCounts2010.2016_TrendAnalysisSubgroup_v3.xlsx
3. Begin using trend analysis program Hannah created. Send any comments or feedback as soon as possible. We will be working with data visualization group to determine how to best display results.
* Overdose Hospitalization 2010-2016-Trend Analysis.sas
Please let me know if you have any questions. Thank you!
Best,
Lia Pizzicato
From: Lia Pizzicato
Sent: Monday, November 6, 2017 5:38 PM
To: hannah.yang@mt.gov<mailto:hannah.yang@mt.gov>; gang.liu@health.ny.gov<mailto:gang.liu@health.ny.gov>; mmaurer@coconino.az.gov<mailto:mmaurer@coconino.az.gov>; sarah.nechuta@tn.gov<mailto:sarah.nechuta@tn.gov>; bretta.jacquemin@doh.nj.gov<mailto:bretta.jacquemin@doh.nj.gov>; melissa.jordan@flhealth.gov<mailto:melissa.jordan@flhealth.gov>; ebrutsch@utah.gov<mailto:ebrutsch@utah.gov>; Lily Tyndall Snow (Lily.TyndallSnow@LA.Gov<mailto:Lily.TyndallSnow@LA.Gov>); mamadou.ndiaye@doh.wa.gov<mailto:mamadou.ndiaye@doh.wa.gov>
Cc: Hal Johnson (hal@hjc-epi.com<mailto:hal@hjc-epi.com>) (hal@hjc-epi.com<mailto:hal@hjc-epi.com>); amypatel@cste.org<mailto:amypatel@cste.org>; Thomas Largo (largot@michigan.gov<mailto:largot@michigan.gov>) (largot@michigan.gov<mailto:largot@michigan.gov>); Pasalic, Emilia Sigrid (CDC/ONDIEH/NCIPC) (kse0@cdc.gov<mailto:kse0@cdc.gov>)
Subject: RE: Trend analysis group updates
Hi everyone,
As promised, here is the program for the trend analysis portion. Below is a description of the various parts of the program from Hannah, and I attached a few figures from Hannah for your reference. I apologize for the long email.
A few notes:
* Bretta ran the program using SAS 9.2, and some of the commands did not work in the older version. Please let me know if this would also be an issue for your jurisdiction as we need to think about whether we need a second program. Bretta is having some trouble with her SAS, but she can send out an alternative code once she resolves the issue for anyone using SAS 9.2. Or, if anyone else would like to volunteer to write a code for an older version of SAS, let me know!!
* Hannah did not write programs to look at the shift by age group and intent statistically as some of the categories are very sparse especially for the heroin poisoning indicator in her jurisdiction (see note #3 in her description). We will have graphs to show the data by age group and intent based on the data provided by jurisdictions for the count portion of the project, but do we as a group want to look at the shifts by age group and intent statistically? Or, do we think looking at the overall indicators (i.e. all drug poisonings, opioid excluding heroin, or heroin) statistically is sufficient?
* We are going to shift the deadline for the results of this analysis to December 15. This will hopefully give us time to address this program and give the data visualization group time to prepare spreadsheets. We should hopefully have spreadsheets out to everyone to fill in their data by November 30.
I'm thinking we might be due for a subgroup call as this analysis is getting a bit more complicated. We can see how the call on Thursday goes and then plan accordingly.
Thanks!
Best,
Lia Pizzicato
Description from Hannah:
1. Forecasting:
* OBJECTIVE: Predict the hypothetical scenario under which the transition had NOT taken place (the "counterfactual" scenario) and compare with the observed trend to evaluate the impact of the transition on drug poisoning counts.
* APPROACH:
* Use combined (not stratified) data set - "hospall_ts"
* Using only the ICD9-CM Era data points, predict the trend line (with 95% CI) forward for 5 quarters (Q4 2015 and Q1-Q4 2016) --> do this for all 3 indicators: all drug poisonings (H1), Non-heroin opioid (H2), and Heroin Poisonings (H3)
* To estimate the observed trends I used PROC AUTOREG
* Montana has a lot of Zeroes in our data, especially for indicator H3. I used PROC GENMOD with Poisson distribution
* See the attached graphs (H1, H2, and H3)
1. Interrupted time series analysis (segmented regression):
· OBJECTIVE: Quantify changes (level and/or slope changes) in Drug Poisoning counts following the ICD-10-CM transition & determine whether they are statistically significant
· APPROACH:
* Look at the combined (not stratified) dataset
· Proposed model:
Yt=β0 +β1(preslope)+β2(intervention) + β3(postslope)
· Yt = expected # of drug poisonings at time=t (H1, H2, and H3)
· Preslope- a continuous variable indicating time from start to beginning of transition --> models the pre-transition trend
· Intervention- a dummy variable coded 0 during icd9-cm and 1 during icd-10-cm (Intervention=1 starting from Q4 2015) --> models the level change attributable to the transition between the time points immediately before and after it
· Postslope- a continuous variable coded 0 up to the last time point in icd-9-cm (Q3 2015) and sequentially from 1 thereafter --> models the post transition trend
· β3- β1 = the change in trend (slope) before vs after (We wouldn't necessarily expect a change in the actual incidence of drug poisoning but maybe due to the differences in ICD 10-cm we may see a slope change??....)
* For montana for H1 I saw a significant level change immediately following the transition:
* There was also a slope change post transition but not statistically significant
Parameter Estimates
Variable
DF
Estimate
Standard
Error
t Value
Approx
Pr > |t|
β0
Intercept
1
298.0474
10.0962
29.52
<.0001
β1
preslope
1
-0.6996
0.7363
-0.95
0.3515
β2
intervention
1
-95.8565
26.3257
-3.64
0.0013
β3
postslope
1
3.1000
7.4075
0.42
0.6793
[cid:image003.jpg@01D34F05.CA159840]
1. Examine data by age group and intent
* OBJECTIVE: determine whether patterns seen in the combined dataset are seen in age stratified or intent stratified data
* I am not sure the best way to look at this. The Age stratified data for montana is VERY sparse especially for indicator H3.
* Here is just a simple graph of Montana data for indicator H1 by Intent, but I'm not sure the direction we want to go with it in terms of analysis
[cid:image004.jpg@01D34F05.CA159840]
From: Lia Pizzicato
Sent: Wednesday, November 1, 2017 9:43 AM
To: hannah.yang@mt.gov<mailto:hannah.yang@mt.gov>; gang.liu@health.ny.gov<mailto:gang.liu@health.ny.gov>; mmaurer@coconino.az.gov<mailto:mmaurer@coconino.az.gov>; sarah.nechuta@tn.gov<mailto:sarah.nechuta@tn.gov>; bretta.jacquemin@doh.nj.gov<mailto:bretta.jacquemin@doh.nj.gov>; melissa.jordan@flhealth.gov<mailto:melissa.jordan@flhealth.gov>; ebrutsch@utah.gov<mailto:ebrutsch@utah.gov>; Lily Tyndall Snow (Lily.TyndallSnow@LA.Gov<mailto:Lily.TyndallSnow@LA.Gov>); mamadou.ndiaye@doh.wa.gov<mailto:mamadou.ndiaye@doh.wa.gov>
Cc: Hal Johnson (hal@hjc-epi.com<mailto:hal@hjc-epi.com>) (hal@hjc-epi.com<mailto:hal@hjc-epi.com>); amypatel@cste.org<mailto:amypatel@cste.org>; Thomas Largo (largot@michigan.gov<mailto:largot@michigan.gov>) (largot@michigan.gov<mailto:largot@michigan.gov>); Pasalic, Emilia Sigrid (CDC/ONDIEH/NCIPC) (kse0@cdc.gov<mailto:kse0@cdc.gov>)
Subject: Trend analysis group updates
Hi everyone,
Happy November! I just wanted to follow-up with a few updates.
Quarterly Hospitalization Counts:
I've attached a new spreadsheet that includes a field to put the number of cases where intent is missing. I'm thinking this should only be an issue with ICD-9-CM, but please fill in the new spreadsheet if possible. Please send Quarterly Hospitalization Counts results by the end of the day on Friday, November 3. If you need more time, please let me know.
Validation Dataset:
I received the completed validation dataset for ICD-10-CM. Attached you will find:
* The validation dataset for ICD-10-CM
* Coding scheme for the dataset
* An excel spreadsheet to enter your results
* CDC's Opioid Overdose Indicator Support Toolkit - see specifically page 7
* Prevention for States guidance on generating morbidity indicators under ICD-10-CM
Please send completed spreadsheets using the validation dataset to Tom Largo (largot@michigan.gov<mailto:largot@michigan.gov>). He will check your answers. The validation dataset group should have a dataset for ICD-9-CM ready within the next week or so.
Trend Analysis Program:
Finally, Hannah and Bretta are finalizing the program for the trend portion of the analysis. It should be available by next week. The Data Visualization group will then use the completed program to create an excel spreadsheet for the results. We will send that out when it is ready.
Thank you and please let me know if you have any questions!
Best,
Lia N. Pizzicato, MPH
SAMHSA/CSTE Applied Epidemiology Fellow
Health Commissioner's Office | Opioid Surveillance, Epidemiology, and Prevention Program
Philadelphia Dept. of Public Health
Community Behavioral Health | Performance Evaluation, Analytics, & Research
Philadelphia Dept. of Behavioral Health and Intellectual disAbility Services
