Discussing Mental Health Disorders and Cancer Treatment Outcomes

A cancer diagnosis is often associated with significant psychological distress and an increased incidence of new mental health disorders (MHDs). In a large-scale analysis of data from the University of California Health system published in Cancer involving 371,897 adult patients, investigators found that 10.6% of patients developed a new MHD within 1 year of their cancer diagnosis. The presence of an early MHD was associated with a higher risk of all-cause mortality, particularly during the first 12 to 35 months (HR, 1.51; 95% CI, 1.47-1.56). These findings underscore the importance of prompt screening and integrated psychological support as a standard component of oncology care.

Following the publication of these results, CancerNetwork® spoke with study investigator Julian Hong, MD, MS, associate professor of radiation oncology in the Baker Computational Health Sciences Institute at the University of California, San Francisco (UCSF), and head of Artificial Intelligence at UCSF Helen Diller Family Comprehensive Cancer Center, about the significance of these results. These are the highlights of that conversation.

CancerNetwork: What was the rationale for this study?

Hong: Some of the inspiration for this portion of work came quite some time ago—probably almost a decade now—when there was an increasing appreciation for how mental health diagnoses can essentially affect how patients undergo cancer therapies. We've had a growing appreciation for how important it is. There have been psycho-oncology programs that have been developed around the country, including here, and that was an inspiration.

How was this study designed?

This is a sequel to a study that we did locally at UCSF. My team focuses on using electronic health records to get information out of standard practice, so we designed this study around trying to ascertain different mental health diagnoses and correlate them with different outcomes. We used a number of approaches to do our best to capture, as accurately as possible, mental health diagnoses and disorders. We had built that study at UCSF, and we wanted to generalize it across our entire UC system. It's a rich source of information and a consolidated place for a number of different major academic centers. Unfortunately, all the data that we have here within the UC system is stored in a single database where all the information is consolidated and reformatted into one data format, which is called the observational medical outcomes partnership [OMOP], which consolidates different electronic health records. It's a rich resource that is still a little bit underutilized; we had some nice uses for it during the COVID-19 pandemic. We used that to extract health record data. It's all de-identified by health record data across the UCs. [We were] trying to identify diagnosis codes, identify interventions—like medications that are done for patients who have MHDs—and then correlate that with different cancer outcomes. The idea was to understand some of the correlations across mental health diagnoses and cancer-related outcomes [and] to understand this across different types of cancers. This is all one step in trying to build a larger study to hopefully, ideally, create strategies to intervene in these situations earlier.

What was the primary finding of this study?

The punchline of the study is that there is a substantial number of people [with cancer] who end up developing a MHD, certainly within the short term. That speaks for what we anecdotally observe, that cancer diagnoses are very stressful points in people's lives. That has profound impacts on therapies. The major finding in the study is that this is correlated with an increase in all-cause mortality, and it's particularly salient in the first few years. As we go through time, it looks like there's some temporal dynamics where some of that impact starts to decrease. This is consistent with some other data in specific disease sites. There are data around certain cancers where MHDs can impact how patients tolerate treatments, which is certainly an expected observation, but this is a way to look at the big picture.

Patients with worse-prognostic diseases had greater odds of developing a new MHD compared with patients with better-prognostic diseases. How does this influence how clinicians could prioritize psychological support across disease types?

Causal relationships are always challenging in observational data. The questions that we would ask are about whether cancers with poor prognoses are the primary cause for some of this increase in both mortality and MHDs, or if there is some triangular or causal relationship. In general, to deliver the best care—just thinking about mental health diagnoses in isolation or as a priority themselves—certainly, [it] is important to get people the right care for the things that are actively going on in whatever scenario they're in. It is fundamentally important to manage mental health diagnoses themselves, but also know that there may be some correlation with other characteristics of their treatment and future outcomes.

Some of our work [involves] trying to identify patients who have depression and anxiety earlier and hopefully getting people to our psycho-oncology team earlier. We're working on some related projects in the primary care setting, too. I'm an oncologist, but I’m fortunate enough to have some great collaborators across other disciplines. We are doing some other work on doing better depression screening, for example, so that we can get people to the right treatments earlier.

Patients with early MHDs who were prescribed psychotropic medications had a higher mortality risk. Is that a marker of severe psychological distress, or could it be connected to other treatment-related factors?

The causality part of it is hard. I, personally, would lean towards the former, which is that it's probably more so a marker of more severe mental health diagnoses. Certainly, it's hard; psychotropic medications are not without their own [adverse] effects and interactions. But it's so hard to tease out the causality there. If you ask me, that is a marker for greater severity. We did that analysis, and a fundamental part of that was that we were trying to come up with different ways to work with imperfect data, which is the case with health records. Looking for medication/prescriptions is at least a little bit more of an objective marker to help us tease some of those things out. We're looking for different ways to analyze the data and help us come up with different looks.

The other part of it is that different psychotropic medications are also used for a number of [other] indications. Some of those things can be markers for other types of treatment, with other specific [adverse] effects of treatment like nausea. There's a lot to disentangle there. That's tricky, but doing a deeper dive into the different medications to understand those issues is important.

To address the future research question, broadly, a big part of our effort is dedicated towards applying large language models and AI to do better information retrieval, which is basically trying to get more content out of our traditional unstructured text—how we describe things in medical notes. For one part, we have a paper that's on the runway, so to speak, looking at applying natural language processing to understanding different elements that are described in medical notes that correlate with interventions. All those things can help us do a better job of teasing out some of these causal questions and severity.

Were there any specific trends in the timing of mental health diagnoses that suggest when a clinician should be most wary or vigilant about screenings?

Really early on. As a clinician, that initial diagnosis is such a tremendously stressful time for patients; that's certainly one time to be very cognizant of it. It's a little bit harder to tease out in a diverse, heterogeneous population, like what we have in this study, but I would expect that any time there are significant changes and things going on—[such as] disease progression, changes in therapies, [adverse] effects—those are times where we should be proactive in thinking about people's mental health. At least from the study, the big spike that you see in the figures is during initial diagnosis. Part of that has to do with how the study was designed.

What are some practical ways to implement mental health screenings into routine workflows without overburdening the already busy staff?

I would say a few things. We've done a better job of routinely screening for things over time. A lot of that is facilitated through patient-reported outcomes. At least in our cancer clinics, there's been a lot of work on trying to give people questionnaires that are oriented towards capturing this type of information. A classic one is the NCCN Distress Thermometer, which is, at its most basic level, a fairly quick questionnaire. We've done some work, even just with a number on a scale from [1 to] 10. It does a great job in identifying a diverse set of things that are correlated with it. That's a reasonably simple metric that can facilitate different forms of triage. The forms of triage are the hard part of that.

Certainly, there are other screening instruments that are out there, [like] the Patient Health Questionnaires [PHQs], for instance. That's a big part of our work in the primary care space. Thinking about ways to streamline that in the oncology space, there's a lot of work being done in that area. In general, with more and more psychiatrists who are focused on oncology, the demand outpaces the supply for these highly specialized teams. [We should] try to come up with ways to help get people to them, deliver that at scale, and to get people in. Those are some of the bottlenecks that are important for us to be thinking about operationally.

As far as a “solution” goes, how can prompt screening mitigate this problem?

That's the key. The patient-reported outcome [PRO] question is a big part of it. What do we do with the information? How do we come up with ways to very rapidly triage people in an appropriate fashion? There's some great work, in general and in the PRO space, with coming up with strategies for intervention. A lot of this is inaccessible to places that have fewer staff and resources. Coming up with good ways to at least speed up or automate triage and referral is very key.

The other hard part with PROs, and they've been implemented differently at different places, is that when folks have PROs assigned to them here at UCSF, they have to complete the questionnaires before they start a telehealth visit. Completion rates of PRO instruments are very challenging. Just to think about how we deliver them electronically, such a big part of the things that we work on here is the digital divide. It’s also challenging to make sure that everybody is getting a chance to voice their needs, to get people to the right place, and to do this in a way that helps everyone.

What future research can build off this study?

The main thing that I'm enthusiastic about is coming up with better ways to get patients to the right place, at the right time. A big part of why we did this study was to set a foundation by doing some detective work to understand exactly what's going on and what we see in the data. Where's the data limited? This series of a couple of studies was intended to help us do that, to set up some of the next steps. But what we're trying to do is to get people to mental health care in a fast, efficient way.

What are we doing next? We're doing work on trying to combine different types of data to identify earlier mental health diagnoses. Even what can feel like small amounts of time—weeks and months—can make a huge difference for people who are going through these conditions. Trying to get people's earlier predictions to facilitate better screening are the cornerstones. Some of that is through data that we collect routinely in our health system. There are also all sorts of other data. From a scientific perspective, one of the things that we're particularly interested in is trying to apply methods around conversations that people are having, whether that's speech patterns, speech characteristics, or the things that people are talking about. We've been doing some early work to see if we can use that to do quicker screening. There are other groups that have worked on that area as well. Then, trying to combine all these different data elements, like PROs or patient-reported questionnaires, and different things in people's healthcare journeys, whether it's cancer or otherwise, to find the people that we need to screen earlier and get them to the right place are all key things that we're trying to tackle. It's one thing to help identify some of these issues and some of these implications of different conditions, but at the end of the day, the goal is to intervene on these things and do a better job of taking care of patients.

Reference

Ganjouei AA, Zack T, Friesner I, et al. Association of mental health disorders and all-cause mortality for patients with cancer: large-scale analysis of University of California Health System Data. Cancer. 2026;132(5):e70254. doi:10.1002/cncr.70254