The goal of AIscreenR is to use AI tools to support screening processes (including title and abstract screening) in systematic reviews and related literature reviews. At the current stage, the main aim of the AIscreenR
package is to support the use and testing of OpenAI’s GPT API models as the second screeners of titles and abstracts or alternatively to reduce research waste, i.e., the number of references needed to be screened by humans. The package allows user to apply OpenAI’s GPT API models from the https://api.openai.com/v1/chat/completions endpoint. In future developments, we expect to add further LLMs when APIs become available for other models, for example from Claude 2. For now, we invite other researchers to test this software so that we as a review community can get a better understanding of the performance of OpenAI’s GPT API models.
Installation
You can install the development version of AIscreenR from GitHub with:
# install.packages("devtools")
devtools::install_github("MikkelVembye/AIscreenR")
Setting API key and checking rate limits
# Find your api key at https://platform.openai.com/account/api-keys
# Thereafter, either encrypt it with the secret functions from the httr2 package
# see https://httr2.r-lib.org/reference/secrets.html or run set_api_key()
# and then enter you key.
library(AIscreenR)
library(revtools)
library(tibble)
library(dplyr)
library(future)
# Setting API
set_api_key(AIscreenR:::testing_key_chatgpt())
# Obtain rate limits info (Default is "gpt-3.5-turbo-0613")
rate_limits <- rate_limits_per_minute()
rate_limits
#> # A tibble: 1 × 3
#> model requests_per_minute tokens_per_minute
#> <chr> <dbl> <dbl>
#> 1 gpt-3.5-turbo-0613 10000 2000000
# Obtain rate limits info across multiple models
rate_limits <- rate_limits_per_minute(model = c("gpt-3.5-turbo-0613", "gpt-4"))
rate_limits
#> # A tibble: 2 × 3
#> model requests_per_minute tokens_per_minute
#> <chr> <dbl> <dbl>
#> 1 gpt-3.5-turbo-0613 10000 2000000
#> 2 gpt-4 10000 300000
How to load ris files. In this example we have downloaded the ris files from the EPPI-Reviewer.
# Loading ris file data via revtools
ris_dat_excl <- revtools::read_bibliography("path/FFT_exclude.ris") |>
suppressWarnings() |>
select(studyid = eppi_id, title, abstract) |>
as_tibble() |>
mutate(
human_code = 0 # Indicating exclusion
)
ris_dat_incl <- revtools::read_bibliography("path/FFT_include.ris") |>
suppressWarnings() |>
select(studyid = eppi_id, title, abstract) |>
as_tibble() |>
mutate(
human_code = 1 # Indicating inclusion
)
filges2015_dat<- bind_rows(ris_dat_excl, ris_dat_incl)
head(filges2015_dat, 10)
#> # A tibble: 10 × 4
#> studyid title abstract human_code
#> <chr> <chr> <chr> <dbl>
#> 1 9434957 Estimating and communicating prognosis in advanc… "Progno… 0
#> 2 9433838 Self-Directed Behavioral Family Intervention: Do… "Behavi… 0
#> 3 9431171 Frequency domain source localization shows state… "The to… 0
#> 4 9433968 A Review of: 'Kearney, C. A. (2010). Helping Chi… "The ar… 0
#> 5 9434460 Topographic differences in the adolescent matura… "STUDY … 0
#> 6 9433554 BOOK REVIEW "The ar… 0
#> 7 9435130 Rapid improvement of depression and quality of l… "Backgr… 0
#> 8 9432040 Pictorial cognitive task solving and dynamics of… "AIMS: … 0
#> 9 9434093 Enhancing the Impact of Parent Training Through … "New an… 0
#> 10 9431505 EEG spectrum as information carrier "Sponta… 0
Example of how to enter a prompt. Can also be done in word (see vignette).
prompt <- "Evaluate the following study based on the selection criteria
for a systematic review on the effects of family-based interventions on drug abuse
reduction for young people in treatment for non-opioid drug use.
A family-based intervention (FFT) is equivalent to a behavior focused
family therapy, where young people’s drug use is understood in relation to family
behavior problems. Family-based interventions also includes manual-based family therapies as
it targets young people and their families as a system throughout treatment, and thereby recognizes
the important role of the family system in the development and treatment of young people’s drug use
problems. FFT was developed in the late 1980s on request from the US National Institute on Drug Abuse
(NIDA). The development of FFT was initially heavily inspired by the alcohol abuse program
Community Reinforcement Approach (CRA), which was aimed at restructuring the environment
to reinforce non-alcohol associated activities. FFT developed to have more emphasis on
contingency contracting, impulse control strategies specific to drug use,
and increased emphasis on involvement of family members in treatment.
FFT is designed to accommodate diverse populations of youths with a variety of behavioral,
cultural and individual preferences. FFT has evolved for use in severe behavioral disturbances
known to co-exist with substance use and dependence, and the core interventions
have been enhanced to address several mental health related problems commonly occurring
as comorbid conditions in drug use treatment participant. For each study,
I would like you to assess: 1) Is the study about a family-based intervention,
such as Functional Family Therapy, Multidimensional Family Therapy, or
Behavioral Family Therapy? (Outpatient manual-based interventions of any
duration delivered to young people and their families). If not, exclude study.
2) Are the participants in outpatient drug treatment primarily
for non-opioid drug use? 3) Are the participants within age 11–21?"
Approximate price of screening
app_obj <-
approximate_price_gpt(
data = filges2015_dat,
prompt = prompt,
studyid = studyid, # indicate the variable with the studyid in the data
title = title, # indicate the variable with the titles in the data
abstract = abstract, # indicate the variable with the abstracts in the data
model = c("gpt-3.5-turbo-0613"),
rep = 10
)
app_obj
#> The approximate price of the (simple) screening will be around $4.477.
app_obj$price_dollar
#> [1] 4.477
app_obj$price_data
#> # A tibble: 1 × 5
#> model iterations input_price_dollar output_price_dollar total_price_dollar
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 gpt-3.5-… 10 4.42 0.0594 4.48
Example of how to conduct simple screening, getting return 1 if a reference should be included, 0 if excluded, and 1.1 if a firm decision cannot be reached.
# Subsetting the number of references to speed up the tutorial screening
plan(multisession)
test_obj <-
tabscreen_gpt(
data = filges2015_dat[c(140:150),],
prompt = prompt,
studyid = studyid, # indicate the variable with the studyid in the data
title = title, # indicate the variable with the titles in the data
abstract = abstract, # indicate the variable with the abstracts in the data
model = c("gpt-3.5-turbo-0613"),
reps = 2 # Number of times the same question is asked to ChatGPT
)
#> * The approximate price of the current (simple) screening will be around $0.036.
plan(sequential)
test_obj
#> Find data with all answers by executing
#> x$answer_data_all
#>
#> Find data with the result aggregated across multiple answers by executing
#> x$answer_data_sum
#>
#> Find total price for the screening by executing
#> x$price_dollar
# Data sets in object
price_dat <- test_obj$price_dat
price_dat
#> # A tibble: 1 × 5
#> model iterations input_price_dollar output_price_dollar price_total_dollar
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 gpt-3.5-… 2 0.0323 0.00048 0.0328
all_dat <- test_obj$answer_data_all
all_dat |> select(human_code, decision_binary)
#> # A tibble: 22 × 2
#> human_code decision_binary
#> <dbl> <dbl>
#> 1 0 0
#> 2 0 0
#> 3 0 0
#> 4 0 0
#> 5 0 0
#> 6 0 0
#> 7 0 0
#> 8 0 0
#> 9 0 0
#> 10 0 0
#> # ℹ 12 more rows
sum_dat <- test_obj$answer_data_sum
sum_dat |> select(human_code, final_decision_gpt:final_decision_gpt_num)
#> # A tibble: 11 × 3
#> human_code final_decision_gpt final_decision_gpt_num
#> <dbl> <chr> <dbl>
#> 1 0 Exclude 0
#> 2 0 Exclude 0
#> 3 0 Exclude 0
#> 4 0 Exclude 0
#> 5 0 Exclude 0
#> 6 0 Exclude 0
#> 7 0 Exclude 0
#> 8 0 Exclude 0
#> 9 0 Exclude 0
#> 10 0 Exclude 0
#> 11 0 Exclude 0