About ha33

ha33 · ‎01-19-2024

Hello everyone, I am attempting to model the effects of an intervention in the same individuals before and after the initiate it. I am not a statistician by training so I might be using some terms wrong. The response variable is whether or not individuals redeem a prescription (binary). Data looks similar to this: ID month half_year outcome prev_treated 1 -3 -1 0 0 1 -2 -1 0 0 1 -1 -1 1 0 1 0 1 1 0 2 1 1 0 1 2 2 1 1 1 Month 0 is the month of the start of intervention. Half_year is 6-month period in relation to start of the intervention. Prev_treated denotes whether individuals had attempted a former treatment. We are interested in estimating the odds ratio for outcome=1 in half_years -4, -3, -2, -1, 1, 2, 3, 4. We expect that: 1) The likelihood of experiencing an outcome at time of intervention is different for individuals. 2) Individuals will have a different change in likeilhood post-intervention (response will be individual). 3) The "measurements" are performed at equally spaced intervals, and closer measurements will be correlated. Considering this, we have made a model with proc glimmix that looks like this: proc glimmix data = have; class ID half_year(ref-4) month prev_treated; model outcome(Event="1")= half_year prev_treated / dist=binary link=logit oddsratio s; random intercept / subject=ID; random month/subject=ID residual type=AR(1); run; We have included the two random statements to accomodate both 1) and 2). I am a little confused as to why month must be in the class-statement to be modeled as a random effect. Also I am not sure that the longitudinal nature of the data is captured (it is not just randomly repeated measurements, but ordered measurements that are correlated) but I guess the type=AR(1) might capture this? I have also read some documentation about G- and R-side effects in relation to the two random statements, but isn't the point to accomodate random variation in starting points (intercepts) between individuals, and random variation in developments/"slope" between individuals after intervention? All in all, I would like some feedback model and if you think it makes sense. Thanks.

ha33 · ‎11-15-2023

Hi everyone, I am a non-statistician looking for some advice on how to intepret the fit statistics in proc glimmix. I am computing odds ratios for an event (0/1) over time in the same individuals. Data looks something like this: ID period outcome season prev_treated 1 1 0 3 0 1 2 0 4 0 1 3 1 1 0 1 4 1 2 0 2 1 0 2 1 2 2 1 3 1 It might be worth adding that the share of observations in which outcome=1 is small, approx 10-20%. The current model looks like this: proc glimmix data = data1 method=rspl plots=oddsratio; class ID period(ref="1") season prev_treated; model outcome(event="1")= period season prev_treated / dist=binary link=logit oddsratio s; random intercept / subject=id; random period/subject=id residual type=AR(1); run; I have two questions : 1) Is there any way, based on this information, to determine which method should be used (RSPL, RMPL, MSPL, MMPL)? 2) In relation to 1), how do I interpret the Fit statistics table: -2 res log pseudolikelihood Generalized Chi-Square Gener. Chi-Square/DF Meaning can it be used like AIC, where lower is better, for example when specifyiung different methods in the method= statement or covariance structures in the type= statement. (AR(1), ARMA (1,1) and TOEP are of interest). Also feel free to comment on the model, if you have other suggestions. Thanks

ha33 · ‎11-09-2023

Hi everyone, I have some individual-level prescription redemption data and I would like to calculate the point prevalence of drug use, defined as the percentage of the population being treated at monthly intervals one year before and after to the initiation of an intervention (intervention starts at different time points for each individual, but every individual has 1 year-follow-up in both directions). My current data looks something like this: Table 1 id drug_atc date_of_redemption grace_period days_supply last_day_of_supply intervention_date cov_by_last intervention 1 AA10A 23/12/2020 30 30 21/02/2021 01/02/2021 - 0 1 AA10A 31/01/2021 30 30 01/04/2021 01/02/2021 1 0 1 AA10A 04/02/2021 30 30 05/04/2021 01/02/2021 1 1 1 AA10A 07/03/2021 30 30 06/05/2021 01/02/2021 1 1 1 AA10A 15/05/2021 30 30 14/07/2021 01/02/2021 0 1 2 AA10A 02/04/2021 30 60 01/07/2021 01/06/2021 - 0 2 AA10A 05/08/2021 30 60 03/11/2021 01/06/2021 0 1 2 AA10A 06/09/2021 30 30 05/11/2021 01/06/2021 1 1 Last_day_of_supply is the days_supply + grace_period. Cov_by_last denotes if the prescription was filled before the last_day_of_supply of the previous prescription (yes=1, no=0). Intervention denotes if the prescription was filled before (0) or after (1) the intervention. I also have a shell table with each id, the date of intervention and date of the monthly intervals since intervention: Table 2 ID drug_atc month_since_int intervention_date intervention date_since_inv 1 AA10A -12 01/02/2021 0 01/02/2020 1 AA10A -11 01/02/2021 0 01/03/2020 1 AA10A -10 01/02/2021 0 01/04/2020 1 AA10A -9 01/02/2021 0 01/05/2020 1 AA10A -8 01/02/2021 0 01/06/2020 1 AA10A -7 01/02/2021 0 01/07/2020 1 AA10A -6 01/02/2021 0 01/08/2020 1 AA10A -5 01/02/2021 0 01/09/2020 1 AA10A -4 01/02/2021 0 01/10/2020 1 AA10A -3 01/02/2021 0 01/11/2020 1 AA10A -2 01/02/2021 0 01/12/2020 1 AA10A -1 01/02/2021 0 01/01/2021 1 AA10A 0 01/02/2021 1 01/02/2021 1 AA10A 1 01/02/2021 1 01/03/2021 1 AA10A 2 01/02/2021 1 01/04/2021 1 AA10A 3 01/02/2021 1 01/05/2021 1 AA10A 4 01/02/2021 1 01/06/2021 1 AA10A 5 01/02/2021 1 01/07/2021 1 AA10A 6 01/02/2021 1 01/08/2021 1 AA10A 7 01/02/2021 1 01/09/2021 1 AA10A 8 01/02/2021 1 01/10/2021 1 AA10A 9 01/02/2021 1 01/11/2021 1 AA10A 10 01/02/2021 1 01/12/2021 1 AA10A 11 01/02/2021 1 01/01/2022 1 AA10A 12 01/02/2021 1 01/02/2022 2 AA10A -12 01/06/2021 0 01/06/2020 2 AA10A -11 01/06/2021 0 01/07/2020 2 AA10A -10 01/06/2021 0 01/08/2020 I would like to create a table like this, table 3 ID drug_atc month_since_int in_treatment intervention 1 AA10A -12 1 0 1 AA10A -11 1 0 1 AA10A -10 1 0 1 AA10A -9 1 0 1 AA10A -8 1 0 1 AA10A -7 1 0 1 AA10A -6 1 0 1 AA10A -5 1 0 1 AA10A -4 1 0 1 AA10A -3 1 0 1 AA10A -2 1 0 1 AA10A -1 1 0 1 AA10A 0 1 1 1 AA10A 1 1 1 1 AA10A 2 1 1 1 AA10A 3 0 1 1 AA10A 4 1 1 1 AA10A 5 0 1 1 AA10A 6 1 1 1 AA10A 7 0 1 1 AA10A 8 1 1 1 AA10A 9 0 1 1 AA10A 10 1 1 1 AA10A 11 0 1 1 AA10A 12 0 1 2 AA10A -12 1 0 2 AA10A -11 1 0 2 AA10A -10 0 0 where in_treatment denotes whether the individual was covered by a days_supply of the drug on the day of the start of the month_since_int. Eventually, the final table should look something like this, which contains the percentage of individuals in treatment at each month in relation to the intervention. month_since_int drug_atc in_treat_percent -12 AA10A 70.0% -11 AA10A 60.0% -10 AA10A 80.0% -9 AA10A 70.0% -8 AA10A 60.0% -7 AA10A 60.0% -6 AA10A 70.0% -5 AA10A 70.0% -4 AA10A 60.0% -3 AA10A 80.0% -2 AA10A 60.0% -1 AA10A 70.0% 0 AA10A 60.0% 1 AA10A 60.0% 2 AA10A 50.0% 3 AA10A 40.0% 4 AA10A 50.0% 5 AA10A 30.0% 6 AA10A 30.0% 7 AA10A 20.0% 8 AA10A 20.0% 9 AA10A 30.0% 10 AA10A 20.0% The key step I am struggling with is creating table 3, but I happy to receive input to any of the steps, it does not have to be the total solution. Thanks

ha33 · ‎10-24-2023

Hi everyone, So I have two datasets, one contains unique IDs, prescriptions and their duration by a start date and end date: Table 1 ID ATC P_DATE LAST_DATE 1 AA10 03_01_12 03_02_12 1 AA10 03_01_12 03_02_12 1 AA10 03_01_12 03_02_12 2 AA10 14_02_12 14_03_12 2 AA10 15_04_12 15_06_12 2 BB10 12_01_12 12_02_12 2 BB10 15_06_12 15_07_12 3 CC10 27_08_13 27_09_13 3 CC10 12_09_13 12_10_13 In another table I have the IDs and start/end dates of months in relation to an intervention. Table2 ID month_num month_num_start month_num_end iv 1 -2 01_02_12 01_03_12 0 1 -1 01_03_12 01_04_12 0 1 1 01_04_12 01_05_12 1 1 2 01_05_12 01_06_12 1 2 -2 15_12_14 15_01_15 0 2 -1 15_01_15 15_02_15 0 2 1 15_02_15 15_03_15 1 I would like to define binary variables (treat_atc) in table 2 that specifies whether the individual was covered by a prescription of the respective drug in that month. Coverage could be defined as >50% of the month being covered by a prescription. So if the individual redeemed a prescription that supplied 30 days starting on 01_03_2012 (P_date) and the month_num_start of a month_num is 14_03_2012, the treat_atc for that drug would be "1" because the individual is covered 31-14=17 days ~55% of that month. I have been playing around with proc sql and different joins but I haven't cracked the code yet. Feel free to suggest a different approach.

ha33 · ‎10-24-2023

Thanks a lot - that works perfectly.

ha33 · ‎10-24-2023

Hi everyone, I have some prescription redemption data that looks something like this: ID ATC P_DATE PACKSIZE DURATION(days) 1 AA10 03_01_12 100 30 1 AA10 03_01_12 100 30 1 AA10 03_01_12 100 30 2 AA10 14_02_12 50 30 2 AA10 15_04_12 50 30 2 BB10 12_01_12 100 60 2 BB10 15_06_12 100 60 3 CC10 27_08_13 50 30 3 CC10 12_09_13 50 30 Here, data are sorted by ID, ATC, P_DATE (ascending). I have the duration in days of each prescription, but some individuals redeem multiple prescriptions for the same drug the same day, and I would like to calculate the total duration for those prescriptions. For example, for the first three rows, I would like to sum the three durations for the prescriptions with the same ID, atc and P_date. My initial idea was to add up the durations using the LAG function in a way such that for the first three rows (same ID, atc and p_date), the durations are added up in one of the three prescriptions (with a duration of 90 days) and then delete the other two using proc sort and nodupkey later. This way the prescriptions become a type of "prescription redemption events" such that for each specific ID, date and ATC code, only one prescription with the summed duration exists. My base code (which unfortunately only partially works - if there are more than 2 identical prescriptiond per date it creates errors and there are other issues): data new; set old; by ID atc p_date; if p_date=lag(p_date) then do; duration=duration+lag(duration); end; run; However, I haven't been able to do this. Does anyone have a suggestion? Feel free to suggest different approaches. Thanks.

ha33 · ‎10-24-2023

Thanks again. About the estimations in the proc gee statement: 'm1' intercept 1 iv 1 0 month 1 iv*month 1 0, 'm2' intercept 1 iv 1 0 month 2 iv*month 2 0, 'm3' intercept 1 iv 1 0 month 3 iv*month 3 0, 'm4' intercept 1 iv 0 1 month 4 iv*month 0 4, 'm5' intercept 1 iv 0 1 month 5 iv*month 0 5, 'm6' intercept 1 iv 0 1 month 6 iv*month 0 6 / e ilink; Should it not be the opposite in terms of iv and iv*month, because iv=1 in months m4-6? In this code, it says iv 1 0 on the months preceding the intervention, and I am wondering it should be the other way around, so that: 'm1' intercept 1 iv 0 1 month 1 iv*month 0 1, 'm2' intercept 1 iv 0 1 month 2 iv*month 0 2, 'm3' intercept 1 iv 0 1 month 3 iv*month 0 3, 'm4' intercept 1 iv 1 0 month 4 iv*month 4 0, 'm5' intercept 1 iv 1 0 month 5 iv*month 5 0, 'm6' intercept 1 iv 1 0 month 6 iv*month 6 0 / e ilink;

ha33 · ‎10-23-2023

Thanks for the clarification. I experience two issues: 1) When I run the NLMeans macro I get the following warning: "The final Hessian matrix is not positive definite, and therefore the estimated covariance matrix is not full rank and may be unreliable. The variance of some parameter estimates is zero or some parameters are linearly related to other parameters." Do you have any idea what the issue could be? The warning appears when some covariates are included (month 1-12 to account for seasonality and previously_treated (binary)) but also when they are not included. 2) My results seem to be the opposite of what I expected (and compared to quite clear trends when just plotting the data). I expected an upward trend in risk before and a downward trend after the intervention when using the NLmeans but I consistently get the opposite. Could there be something in the code that mixes up pre- and post slopes?

ha33 · ‎10-19-2023

Thanks for replying. It runs SAS EG Version 8.3 update 7, so I guess the margins macro isn't included there (yes, it says invocation of the macro is not resolved) . I can't update it or download the macro as it is a government server without internet connection and I don't have administrator rights. However the NLMEANS macro is available and gives the slopes in the 3 month pre and post periods., which is great. I have several more months than I have specified in the table above. I am not quite sure what the all the values in c refer to: pre slope, -.5, 0, .5, 0, 0, 0 post slope, 0, 0, 0, -.5, 0, .5 gap m4-m3, 0, 0, -1, 1, 0, 0 What do these numbers refer to, and how would you adjust if there were e.g. 6 months on each side of the intervention instead of 3?

ha33 · ‎10-18-2023

Thanks again. I run SAS in a protected environment on a remote desktop where the data is stored (GDPR rules) so I don't have the option to download the macro. Event probability would be very useful, but slopes before and after iv and the immediate change in terms of odds ratio is also very informative. Is there any other way of getting an event probability or test the difference in slope before and after without using the macro?

ha33 · ‎10-17-2023

Thanks a lot! The GEE model works but the margins macro doesn't (I work in enterprise guide, maybe that's why) Anyway that was very helpful.

ha33 · ‎10-13-2023

Hi everyone I want to perform an interrupted time series -like analysis where we assess the effect of an intervention on experiencing an event. It is observational data that looks something like this: ID MONTHS INTERVENTION OUTCOME 1 1 0 0 1 2 0 0 1 3 1 0 1 4 1 1 2 1 0 0 2 2 0 0 2 3 1 1 If the outcome were continuos I would have used a mixed model beautifully described in an article form SAS global forum 2020: PROC MIXED DATA = SINGLE_ITS_DATA METHOD=ML PLOTS(MAXPOINTS=20000)=(RESIDUALPANEL(UNPACK) VCIRYPANEL(UNPACK) ); CLASS ID; MODEL OUTCOME=MONTH INTERVENTION MONTH*INTERVENTION/S; REPEATED / SUBJECT = ID TYPE = UN; ESTIMATE 'PRE-PERIOD SLOPE' MONTH 1 / CL E; ESTIMATE 'POST-PERIOD SLOPE' MONTH 1 MONTH*INTERVENTION 1 / CL E; ESTIMATE 'PRE/POST GAP' MONTH 1 INTERVENTION 1 MONTH*INTERVENTION 10 /CL E; I want to perform the same analysis but since my outcome is binary I was thinking of proc genmod or proc glimmix, and estimating the odds ratio of experiencing the effect in the period before, after and whether the difference between before and after is significant. Anyone who could suggest a syntax for this that comes close to what the mixed model delivers above? Doesn't matter which procedure.. Best regards, Kristian

ha33 · ‎02-13-2023

Hey so an update: I ended up using PROC SURVEYSELECT and defining each patient as one stratum. Worked perfectly.

ha33 · ‎02-07-2023

Thanks for you reply. There is no rule as everyone in the background population is matched by the same variables. I will attempt your suggestion.

ha33 · ‎02-07-2023

Hi, I have the following code joining two tables by an unique identifier, in which a disease population is joined with a background population, creating approx 1000 matches per person in the disease population. proc sql; create table new as select distinct case_id, id from disease_pop inner join background_pop on disease_pop.case_id = background_pop.id; This makes a table that looks like this: CASE_ID ID 123 345 123 567 123 891 234 902 234 903 234 904 Only there are approximately 1000 "IDs" for each CASE ID So what I want to do next is reduce the number of matches, so instead of having 1000 IDs per Case ID I would like only 10 IDs per CASE ID, but I still want to have all CASE IDs in my new table. Any suggestions?

Online Status	Offline
Date Last Visited	‎04-11-2025 04:50 AM

Random statements in a repeated measurements logistic model using glim...

Interpreting fit statistics in proc glimmix with a binary outcome and ...

Calculating treatment adherence/persistence from individual level drug...

How to compute binary variable indicating treatment status by overlapp...

Re: Summing rows with identical values using the LAG function and keep...

Summing rows with identical values using the LAG function and keeping ...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Random statements in a repeated measurements logistic model using ...

Re: How to assign a Count of dataset to variable and use it in IF ELSE...

Re: Error using macro variable in table column name in inner join

Pass Date macro Variable into a PROC SQL Where statement

Re: Pass Date macro Variable into a PROC SQL Where statement

Random statements in a repeated measurements logistic model using glim...

Interpreting fit statistics in proc glimmix with a binary outcome and ...

Calculating treatment adherence/persistence from individual level drug...

How to compute binary variable indicating treatment status by overlapp...

Re: Summing rows with identical values using the LAG function and keep...

Summing rows with identical values using the LAG function and keeping ...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Re: Interrupted time series using prox glimmix or proc genmod with a b...

Interrupted time series using prox glimmix or proc genmod with a binar...

Re: Selecting fixed number of matches in an inner join

Re: Selecting fixed number of matches in an inner join

Selecting fixed number of matches in an inner join