Happy New Year! My first post in the new year was also recently featured at Rehabs.com, and it talks about research from our group.

Direct Cash transfer is a very trendy. See this, this, and most recently this, complete with the associated theatrics of a Chinese billionaire attempting and failing to give $300 to every homeless man and woman who shows up to a luncheon. There are not many rigorously conducted randomized controlled trials on Direct Cash Transfer, but Johannes Haushofer and colleagues’ trial in Kenya (table 4) provides some convincing evidence that cash reduced consumption of alcohol and tobacco and increased consumption of meat and fish--a pretty impressive outcome.

Treating people in poverty with cash “works," on average, in improving psychological well-being and reducing substance use, but there is resistance in implementation. Says Christopher Blattmann in the New York Times: 

“The executive director of Rescue Mission said he was worried that people might spend the handout on drugs or alcohol. This pessimism (and paternalism) is common and understandable.”

The detractors have a point: if you work in this field, you know that some people do spend their cash on drugs or alcohol, even though most of them do not. Why does it only work in some people but not in others, and how can a “Big Data” approach help?  

Direct cash transfer has two forms: conditional and unconditional. Unconditional cash transfer is really a form of conditional cash transfer with a very low barrier condition to get the cash: all that is required is attendance and tracking as part of the program. When we limit our outcome of interest to only substance abuse recovery, there is a large literature around conditional cash transfer, better known as contingency management. This method is pioneered by Stephen Higgins [1] in treating cocaine in a very specialized context called Community Reinforcement Approach (CRA) with voucher reward, and subsequently used in treating a variety of substance use disorders, including opiates, alcohol and marijuana [2], often achieving dramatic results in randomized controlled trials.  The treatment effect of this kind of psychotherapy, which incentivizes negative urine toxicology screen with sometimes escalating amounts of cash, compares very favorably to the best medication treatment of substance use disorders. In a typical study like this for cocaine use disorder, in the treatment (cash reward) group, about half (50%) of the patients are able to achieve some positive outcome (i.e. abstinence), whereas in the control (waiting-list) group, only about 10% are able to do so. This is roughly as good as treating opiate dependence with methadone, one of the most effective pharmacologic treatments in all of medicine. This suggests that cash transfer is in fact one of the most effect interventions for substance abuse.  

But wait a second, what happens to the other 50% who do not respond to the cash incentive? This is where things get tricky. When people look for predictors of treatment response, like gender, socioeconomic status and psychiatric disorders, sometimes they can find an effect here and there, but the results are difficult to interpret: perhaps there is a statistically significant correlation between a predictor and outcome, but the effect size is so small that it is practically useless on an individual basis. We do not know for any particular individual if cash incentive would work. We certainly do not understand the underlying mechanism of individual variations in treatment response.  

The persistence of the addictive behavior despite strong financial incentive in this group of individuals (“the other 50%”) and our curiosity for possible mechanisms make us conjecture that there is a genuine brain circuit level abnormality. Diana Martinez and colleagues at Columbia demonstrated a measurable decrease in dopamine release in a specific part of the brain, called the striatum, for patients who did not respond to cash incentives [3] compared with those who did. I became interested in whether this signal would be helpful in predicting of treatment response on an individual level. Collaborating with Diana’s group and a number of other investigators in our division, and switching our ways of thinking from testing hypotheses to building predictive models, we applied elementary machine learning techniques to develop and evaluate predictive models using a combination of this brain derived signal and early treatment response. Surprisingly, we can get >90% accuracy in predicting who will be a treatment responder and who will not on an individual basis [4].  There are many caveats: complex model comparison, cross validation and evaluation procedures are not as rigorous as they can be due to small sample size. Nevertheless, our work raises the tantalizing possibility that we may be closer than we think in figuring out who will and who will not respond to cash incentives, and therefore be able to customize and refine treatment based on a better understanding of these mechanisms at the level of neural circuits. 

Cash incentive is not a panacea: it does work, on average, better than no treatment. An appropriately designed psychotherapy with cash incentive can be as effective as any medical intervention in altering human behavior. In fact, it may be an effective strategy to alter human behavior beyond medicine. However, cash incentive does not work for everyone, and maybe it does not work when the part of the brain responsible for rational judgment and prioritizing of reward becomes dysfunctional, either due to chronic drug use, or due to some other external stressor. If we understand this neural circuit abnormality better, we can identify “the other 50%” and change their brain with either medication or some other form of psychotherapy, and achieve an even better treatment response. Predictive modeling of individual response to Direct Cash Transfer is one step towards personalized medicine for substance use and other associated disorders.

References

[1] Higgins, S.T., Delaney, D.D., Budney, A.J., Bickel, W.K., Hughes, J.R., Foerg, F., & Fenwick, J.W. (1991). A behavioral approach to achieving initial cocaine abstinence. American Journal of Psychiatry, 148 (9), 1218-1224.

[2] Lussier, JP, et al. (2006). A meta-analysis of voucher-based reinforcement therapy for substance use disorders. Addiction, 101, 192–203.

[3] Martinez et al, (2011) Imaging dopamine transmission in cocaine dependence: response to treatment linked to neurochemistry Am J Psychiatry. 168(6): 634–641.

[4] Luo et al, (2014) Multimodal predictive modeling of individual treatment outcome in cocaine dependence with combined neuroimaging and behavioral predictors. Drug Alcohol Depend. 2014 Oct 1;143:29-35.

About the Author

Sean X. Luo M.D., Ph.D.

Sean X. Luo, M.D., Ph.D., is a physician-scientist working at Columbia University and The New York State Psychiatric Institute.

You are reading

Hooked on Patterns

Online Therapy: How Does This Work?

A review of this rapidly moving, possibly dystopic technology

Understanding Addiction With Electronic Registries

Disruptive technology with a $40 per patient clinical trial price tag.

Optimal Time for Fund Raising and Grant Writing

How many hours should I spend on a grant? A mathematical analysis.