Verbatim abstract text in blue follows each analysis.
SUBMITTED Version of ICPE 2020 KY Survey Early Prescribers (2339 characters of 2350 limit)
Do Early Prescribers Of New Drugs Have Different Risk Management Practices?
Submitted ICPE 2020 abstract #4157
Dasgupta N, Brown JR, Freeman P, Nocera M, Slavova S
Uptake of new technology occurs first in "early adopters." We hypothesized that early prescribers of new medicines could likewise be identified, possibly influencing interpretation of early post-market assessments.
Conduct a statewide prescriber survey to evaluate if self-described early prescribers have differences in risk management practices compared to later prescribers.
Planned hypothesis generating analysis of a larger survey on opioid prescribing, fielded by state licensing authority in Nov 2019.
Setting: All DEA-registered Kentucky (US) physicians.
Exposure: Self-reported affirmation of any of 3 statements: I prescribe new medications before others; I enjoy the variety of prescribing new medicines; I like to share with colleagues about new medicines I've prescribed. (Other options: I feel more comfortable using familiar medications, etc.).
Outcome: Physician characteristics, opioid prescribing.
Analysis: Descriptive univariable/bivariable stats
Emails were delivered to 7631 physicians, with 651 respondents (8.5% - similar to other unincentivized physician surveys). After limiting to controlled substance prescribers and removing incommpletes, the analysis sample was n=349, with 83 (24%) early prescribers. Early adopters were disproportionately in early or late career (less than 15 or 35+ practice-years). The male:female ratio was 2:1 among respondents, but proportionately more females (26%) than males (19%) were early adopters. There were no differences in patient load, practice setting, or medical specialty except emergency medicine and general surgery were less likely; and oncology and OB/GYN were more likely to be early adopters.
Early prescribers were more likely to use opioid risk stratification tools (OR 1.5; 0.85, 2.7). Abuse deterrent formulation (ADF) opioid prescribing was similar (OR 1.1; 0.65, 1.8). However, early adopters were more likely (OR 1.4; 0.73, 2.6) to prescribe newer non-OxyContin ADFs, with 93% endorsing "innovative nature of abuse-deterrence mechanisms" as a consideration. They supported legislation mandating insurance coverage for ADFs (OR 3.5; 1.6, 8.9), and more strongly endorsed technical solutions like electronic prescription monitoring programs and urine drug screens, while opposing prescribing limits.
In addition to prescribing newer medications, early adopters may differ in other healthcare delivery aspects that may impact care. The survey is being replicated in other states. Quantitative assessments will follow. Studies with historical controls may want to consider impacts of early adopters.
display "Notebook generated on $S_DATE at $S_TIME ET"
cd "/Users/nabarun/Dropbox/Projects/P1 Surveys"
clear all
set more off
frame create pharm
frame pharm: use pharmacistky
frame create docs
frame docs: use prescriberky
frame dir
frame change docs
di "Exclude non-controlled substance prescribers and incomplete surveys:"
keep if contr_sub==1
* n=206 deleted
// Create composite variables for ADF prescribing
local adfs "embeda hysinglaer morphabonder xtampzaer oxycontin"
foreach i of local adfs {
gen any`i'=0
replace any`i'=1 if inlist(`i',2,3,4,5)
order any`i', a(`i')
*tab any`i'
la var any`i' "Any reported prescribing of `i'"
}
gen anyadf=0
replace anyadf=1 if anyembeda==1 | anyhysinglaer==1 | anymorphabond==1 | anyxtampzaer==1 | anyoxycontin==1
la var anyadf "Any ADF prescribing"
tab anyadf
gen anynotocadf=0
replace anynotocadf=1 if anyembeda==1 | anyhysinglaer==1 | anymorphabond==1 | anyxtampzaer==1
la var anynotocadf "Any non-OxyContin ADF prescribing"
tab anynotocadf
// Recode into innovation influence as dichotmous
gen innovinfluence = 0
replace innovinfluence = 1 if inlist(innovate,3,4)
replace innovinfluence=. if innovate==.
la var early_adopt___1 "Usually prescribe new medications before others do"
la var early_adopt___2 "Prefer medications which have worked well for patients in the past"
la var early_adopt___3 "Like being able to share with colleagues about new medications I've prescribed"
la var early_adopt___4 "Like the variety of prescribing new medications"
la var early_adopt___5 "Feel more comfortable using familiar medications"
la var early_adopt___6 "Prefer to wait until I hear about colleagues' experiences with prescribing new medications"
la var early_adopt___7 "Other"
foreach var of varlist early_adopt___1-early_adopt___6 {
tab `var',
}
set linesize 90
list early_adopt_other if early_adopt_other!=""
Self-reported affirmative response to any of 3 statements:
early_adopt___1
prescribing new medications before others
early_adopt___3
enjoying variety of prescribing new medicines
early_adopt___4
liking to share with colleagues about new medicines
// Create variables for early adopters
gen earliest=0
la var earliest "Self-identified early adopters"
replace earliest = 1 if early_adopt___1==1
tab earliest
gen early=0
la var early "Enjoys novel prescribing"
replace early = 1 if early_adopt___1==1 | early_adopt___3==1 | early_adopt___4==1
tab early
Keep if survey is eventually incomplete but respondent had self-identified as an early prescriber or not.
di "Exclude incomplete surveys while retaining early prescribers (see Section 5 for rationale):"
keep if physician_survey_complete==2 | early==1
* n=131 should be deleted
tab earliest
tab early
tab gender early_adopt___1, col nokey
tab gender early_adopt___3, col nokey
tab gender early_adopt___4, col nokey
tab gender earliest, col nokey
tab gender early, col nokey
tab gender early, row nokey
The male:female ratio was 2:1 among respondents, but proportionately more females (26%) than males (19%) were early adopters.
di "Among non-early prescribers:"
tab years_prac if early==0, p
di "Among early prescribers:"
tab years_prac if early==1, p
tab years_prac early, col nokey
ranksum years_prac, by(early)
tab years_prac earliest, col nokey
ranksum years_prac, by(earliest)
There is some murkiness here and low cell sizes, but the trend is worth keeping an eye on.
Early prescribers tended to be disproportionately in early or late career (<15 and 35+ practice-years, respectively).
tab no_patients early, col nokey
There was no difference in number of patients seen per week.
tab specialty_type early, col nokey
There were no significant differences in medical specialty... possibly except emergency medicine and general surgery less likely, and hem/onc and OB/GYN more likely to be early adopters, but cell sizes were small.
tab prac_type early, col nokey
Likewise, there were no appreciable differences in practice setting, possibly with the exception of ED physicians less likely to prescribe new medicines.
tab riskstrat early, m col nokey
gen risktool = 1
replace risktool=0 if riskstrat==.
cc risktool early
Early prescribers were more likely to use opioid risk stratification tools (32% vs. 24%). Early prescribers were more likely to use opioid risk stratification tools (OR 1.5; 0.85, 2.7).
tab anyadf early, col nokey
cc anyadf early
tab anynotocadf early, col nokey
cc anynotocadf early
tab innovate early, m col nokey
ranksum innovate if innovate!=., by(early)
tab innovinfluence early, col nokey
Early and standard prescribers had similar abuse deterrent formulation (ADF) opioid prescribing (OR 1.1; 0.6, 1.8). However, early adopters were more likely (OR 1.4; 0.7, 2.6) to prescribe newer non-OxyContin ADFs, with 93% endorsing "innovative nature of abuse-deterrence mechanisms" as a consideration.
tab state_legis early, col nokey
mdesc state_legis
cc state_legis early
tab dslimits early, col nokey m
ranksum dslimits if dslimits!=5, by(early)
tab urineds early, col nokey m
ranksum urineds if urineds!=5, by(early)
tab kasper early, col nokey m
ranksum kasper if kasper!=5, by(early)
They were more likely to support legislation mandating insurance coverage for ADFs (OR 1.3), and more strongly endorsed technical solutions like electronic prescription monitoring programs and urine drug screens, while more vehemently opposing prescribing limits.
Objective: To assess furthest question responded to as a marker for inclusion. This was deemed important because some respondents clicked on no more than the initial screen.
// Early adopter analysis for IPCE abstract
//
cd "/Users/nabarun/Dropbox/Projects/P1 Surveys"
clear all
set more off
use prescriberky
di "Exclude non-controlled substance prescribers and incomplete surveys:"
keep if contr_sub==1
distinct record_id
di "Check within early adopters:"
gen early=0
la var early "Enjoys novel prescribing"
replace early = 1 if early_adopt___1==1 | early_adopt___3==1 | early_adopt___4==1
tab early
foreach x of varlist _all {
rename `x' zz`x'
}
rename zz* v#, renumber
rename v1 id
rename v106 early
gen incomplete=regexm(lower(v3),"not completed")
di "Drop free text strings:""
drop v2 v4 v3 v12 v13 v21 v33 v63 v65 v73 v81 v82 v98 v101 v104
reshape long v, i(id)
drop if v==. & _j!=2
drop if inlist(v,0,1,2)
bysort id: egen last = max(_j)
drop _j v
duplicates drop
di "Describe length of completeness among incomplete surveys:"
tab last if incomplete==1
di "Look at distribution within early prescribers:"
tab last if incomplete==1 & early==1
About 70% of incomplete surveys respondents stopped after about the 11th question, which was the radio button question about familiarity with ADF formulations and route of preventing abuse. This suggests a general unwillingness to engage in the route of exposure question, despite this being a key element of labeling.
The analytic implication is to not limit by overall incomplete surveys, but rather to limit to those who we have a determiniation of early adopter status. Incomplete responses to specific later survey questions are to be treated as missing. Among early prescribers, there doesn't appear to be much of a pattern...5 or so did not complete beyond the familiarity/route table. Since the early adopter question construction relies on negative data (e.g., boxes unchecked, the handling of missingness can be differential to reflect the variable construction definition.
The analysis above has been updated to refelct this design decision. There are 2 final exclusion criteria:
First, if limiting to controlled substance prescribers, removing 206 observations out of the original 686 responses received:
keep if contr_sub==1
Second, remove incomplete surveys but retain surveys among early prescribers, removing 131 observations for an analysis set of 349.
keep if physician_survey_complete==2 | early==1
In the next iteration of the survey we should think about moving that radio button table or simplifying it.
Replication in other states will be helpful to have more sample size. Further characterization of early adopters could require more or better questions. Using the earliest
self-identified category could serve as a sub-group analysis for a specificity and strength gradient. I wonder if there are measures of sociality (professional connectedness) that might explain some of the tendency to prescribe early, e.g., network effect for incoming information about new drugs or social milleu where having conversations about new drugs is rewarding.
fin.