By Jordan Loewen-Colón May 26th 2023
Psychedelic research labs across the country are deeply engaged in groundbreaking work involving the clinical trials of psychedelic-assisted therapies. These trials seek to gauge the safety and effectiveness of various psychedelic substances, including psilocybin, MDMA, LSD, Ketamine, and Cannabis. We were tasked with developing a predictive model using personality traits and focusing on 'precision' as the key performance indicator, aiming to minimize false positives for identifying potential trial participants. Many of these trails prefer a focused approach, favoring a smaller, more reliable group of participants who are genuinely likely to experiment with psychedelics rather than a larger group with unpredictable inclinations.
Our findings suggest that clinical trails should prioritize incorporating Oscore assessment into screening processes to improve predictions of psychedelic use, as higher scores often indicate an inclination towards such usage. Given the noticeable difference in average Oscores between psychedelic users (0.152) and non-users (-0.593), investigating Oscore's influence on the therapeutic effects of psychedelic-assisted therapies could yield valuable insights. And finally, the study should target those who've never used legal highs, nicotine, or cocaine as potential participants for psychedelic trials.
To make our recommendations, we analyzed the Drug Consumptions (UCI) from Kaggle. As stated on the original database:
- 1885 respondents
- 12 attributes per respondent
- Personality measurements which include: neuroticism (Nscore), extraversion (Escore), openness to experience (Oscore), agreeableness (Ascore), conscientiousness (Cscore), impulsivity (Impulsive), and sensation seeking (SS)
- Demographics like level of education, age, gender, country of residence, and ethnicity.
- 18 legal and illegal drugs: alcohol, amphetamines, amyl nitrite, benzodiazepine, cannabis, chocolate, cocaine, caffeine, crack, ecstasy, heroin, ketamine, legal highs, LSD, methadone, mushrooms, nicotine and volatile substance abuse and one fictitious drug (Semeron) which was introduced to identify over-claimers.
- "For each drug, they have to select one of the answers: never used the drug, used it over a decade ago, or in the last decade, year, month, week, or day."
In technical terms, the dataset has 1884 rows and 31 columns with a mix of floats and objects. The personality scores (rows 6:12) are measured on a Likert-based scale ranging from 0 (“Strongly Disagree”) to 4 (“Strongly Agree”) and then rendered as a float. The demographics have various subcategories, and the drug values are measured by recency (if ever) the substance has been consumed, CL0 being never used, and CL6 being used in the last day. Since we were primarily focused on understanding psychedelic use based on personality scores, we used violin plots to get a sense of the connection.
Since the data contained no null values or duplicates, we were able to focus on creating a target column of just psychedelic drugs: cannabis, ecstasy, ketamine, LSD, and mushrooms. We then dropped any rows that indicated used of the fictious drug "Semer" that the original data collectors had included to weed out "overclaimers." Finally, we created a pipeline to streamline our model production going forward and split the data into training and test sets. Given the imbalance of the data set that was already such a small size, we incorporated SMOTE into the pipeline as part of the preprocessing step to increase fairness across our models.
Our first model was a simple logistic regression. Starting with a logistic regression model offers interpretability and simplicity, serving as an efficient method to establish baseline performance for binary classification, such as distinguishing participants willing to try psychedelics. Its probabilistic output and capability to highlight feature importance provides crucial insights into factors influencing willingness to participate in the trial, while setting a comparative standard for future, more complex models. The model resulted in a precision score of 97%, which implies a lower rate of false positives, as precision is the ratio of true positives to the sum of true positives and false positives.
The model had nine false positives, which isn't bad. We then looked to see what the coefficients with the highest magnitude were:
The Oscore had the largest coefficient magnitude of all our personality traits. The coefficient value of 0.53 for "Oscore" means that for every one-unit increase, the log odds of the outcome "Psychedelics" being 'yes' (versus 'no') increase by 0.5, assuming all other variables in the model are held constant. To better understand this in terms of odds (rather than log odds), we can calculate the odds by taking the exponent of the coefficient: exp(0.5) ≈ 1.65. This means that for every one-unit increase in "Oscore", the odds of the outcome "Psychedelics" being 'yes' (versus 'no') increase by about 65%, assuming all other variables in the model are held constant. And since, as we saw above, people who have taken psychedelics have a higher Oscore, we can assume that a higher Oscore means a higher likelihood that a person has consumed a psychedelic (or perhaps will).
We continued with our modeling by using both a Random Tree Classifier (RFC) and a Gradient Boost classifier (GBC), which produced precision ratings of 97% as well. Looking at the feature importances of both revealed an agreement between the Log and GBC models that Oscore is the most important feature, while our RFC model thought Oscore was second to SS.
RFC Feature Importances:
GBC Feature Importances:
When comparing all our models, it looks like our Logistical Regression model scores highest on accuracy and F1. The RFC model scored highest on precision and recall. While the scores are close, we'll give the Log model the edge and choose it to draw understandings.
Model: Logistic Regression
- accuracy: 0.8763326226012793
- precision: 0.9723926380368099
- recall: 0.8661202185792349
- F1-score: 0.9161849710982659
Model: RFC
- accuracy: 0.8571428571428571
- precision: 0.9746031746031746
- recall: 0.8387978142076503
- F1-score: 0.9016152716593245
Model: GBC
- accuracy: 0.8528784648187633
- precision: 0.9684542586750788
- recall: 0.8387978142076503
- F1-score: 0.8989751098096632
Interpretations:
Out of the 130 coefficients, our Oscore is in the top 35, but there is a significant difference between it and our leading coefficients: Never Having Taken a Legal Highs, Nicotine, and Cocaine. A question to ask is, do people who take psychedelics and have NEVER taken a Legal High score higher on openness than non-psychedelic consumers?
As it turns out: The average Oscore for individuals who scored 'CL0' in the Legalh column and are categorized as 'Psychedelics' is: -0.13 The average Oscore for individuals who scored 'CL0' in the Legalh column and are categorized as non-psychedelics is: -0.6 The average Oscore for all other results in the Legalh column is: 0.41
Looks like psychedelic users score higher on the Oscore than non-psychedelic users! So that indicates that Oscore possibly contributes positively to Psychedelic use in conjunction with our most important coefficient. However, it looks like the average Legal High users scores even higher. That means we will want to filter them out. But what about Oscore more generally?
The average 'Oscore' for individuals who take psychedelics is: 0.15 The average 'Oscore' for individuals who do not take psychedelics is: -0.59
Our first recommendation involves Recruitment Strategy. The precision of 97% achieved by the logistic regression model indicates that the model is effective in identifying potential trial participants who are genuinely likely to experiment with psychedelics. The institute can focus on targeting individuals who exhibit characteristics associated with high precision, such as never having taken legal highs, nicotine, or cocaine. These factors can be used as screening criteria during the recruitment process.
Our second recommendation involves the importance of the Oscore. The Oscore coefficient with a magnitude of 0.5 compared to the other personality traits indicates that it is one of the significant predictors of psychedelic use. This model indicates that individuals with higher Oscores tend to be more inclined toward using psychedelics. Therefore, considering an individual's Oscore can contribute positively to the prediction of psychedelic usage. The institute can incorporate the assessment of Oscore into the screening process to further refine the selection of potential participants.
Our final recommendation involves a comparison of Oscore and psychedelic use. The analysis of the average Oscore for individuals who take psychedelics and those who do not reveals a notable difference. Individuals who take psychedelics have an average Oscore of 0.152, while those who do not have an average Oscore of -0.593. This indicates that Oscore may be a relevant factor in understanding the inclination towards psychedelic use. The institute can explore further research to investigate the relationship between Oscore and the therapeutic effects of psychedelic-assisted therapies.
Our logistic regression model may have scored so high for two reasons:
-
Because log regressions assume that there's a linear decision boundary between the classes, while decision trees (and by extension, Random Forests and Gradient Boosting models) do not. If the data indeed has a linear decision boundary, logistic regression might outperform more complex models.
-
Random Forest and Gradient Boosting models are more complex than logistic regression, and this complexity can lead them to overfit the training data, especially if the dataset is small, which ours is.
Therefore, getting more data might actually allow our more complex models to provide more precise predictions.
For a full analysis, please check the Jupyter Notebook or slide presentation. Further questions? Contact Jordan Loewen-Colón @ [email protected]
├── data : data used for modeling ├── images : images used in PPT and README ├── draft 1 : previous files from first draft of project ├── SMOTE version.ipynb : notebook used to pull from API ├── README.md : project information and repository structure ├── presentation.pdf : the PowerPoint presentation used to present data analysis