The timing of being asked to write this blog coincided with the publication of the draft for consultation on NICE Guidance regarding ‘Depression in Adults: treatment and management’ (NICE, 2021). For anyone who has read the consultation document and especially for those working in IAPT (Improving Access to Psychological Therapy) services, there have been many concerns raised about the recommendations within it. The BABCP has issued a response running to almost sixty pages, which indicates the extent of these concerns (BABCP, 2022).
IAPT services use a stepped care model meaning that where it is indicated from the evidence base, patients with certain presenting issues (including depression) are likely to move through different steps of treatment intervention. This usually starts at step 2 for patients with a first episode of depression or mild to moderate symptoms and would consist of an intervention such as 6-8 sessions of guided self-help with a Psychological Wellbeing Practitioner (PWP) and may progress to a step 3 intervention such as 10-20 sessions of cognitive behavioural therapy (CBT), counselling for depression or couples counselling if required. Many IAPT services will facilitate patients with more complex presentations, such as recurrent depression or more complex additional issues to begin at step 3 of the model as appropriate.
Within the consultation is the implication that the stepped care model should be removed and that patients presenting with a new episode of depression should move directly to a high-intensity intervention rather than receive low-intensity interventions usually recommended as a first line treatment. Practical, staffing/training and financial implications of this aside, the issue that most patients and clinicians want to answer remains – is this more clinically effective?
The current paper by Delgadillo et al. (2021) aims to examine whether stratified care, using a personalised treatment recommendation for each patient based on a machine learning algorithm, using information based on demographics and presenting issues, was more effective than stepped care.
The authors describe their study design as a ‘pragmatic, multisite, single-blind, cluster randomised clinical trial’. Their hypothesis is that stratified care will lead to improved rates of recovery for patients with depression as compared to those receiving stepped care due to patients with more complex case presentations receiving a high-intensity treatment earlier.
The authors used a sample of 30 clinicians (qualified PWPs) from four different services, who during an 18 month period, assessed 951 patients meeting criteria for the study:
- had a presenting issue that is within the IAPT remit to treat (IAPT Manual, 2020)
- accessed treatment, defined as attending at least one session after their assessment
- were not receiving treatment elsewhere at the time of assessment
Assessing PWPs were randomly assigned to either the stepped care or stratified care condition and they were made aware of their allocation. All used the same assessment template with patients.
Patients allocated to the stratified care condition had their details analysed by an artificial intelligence application, using measures of depression, anxiety, functional impairment, personality traits, employment status, race and ethnicity. The machine learning algorithm then made a personalised treatment recommendation based on whether a patient’s case was deemed to be ‘standard’ or ‘complex’ as defined by the probability of that patient reaching clinical recovery from depression and anxiety symptoms after treatment. The approach aimed to fast track patients with ‘complex’ presentations and those with conditions for which psychotherapy is indicated straight to step 3.
After allocation, patients received the clinical intervention appropriate to that decision in their local service with the first available and appropriately trained clinician. Using a pragmatic design, the interventions provided were not modified or monitored in any way to ensure the offer was consistent with routinely delivered care.
The primary outcome measure used was the Patient Health Questionnaire (PHQ-9, Kroenke et al, 2001) which is the standard measure used in IAPT services to monitor response to treatment for depression. The Generalised Anxiety Disorder Questionnaire (GAD-7, Spitzer et al, 2006) was used as a secondary outcome measure.
Of 1,453 patients screened in the study period, 951 met eligibility criteria to participate. 583 were allocated to stratified care and 368 to stepped care. Approximately two thirds of the sample were female with mean age of 38.66 in the stratified care group and 37.65 in the stepped care group. Over 95% of the sample identified as White.
The stratified care pathway allocated 50.9% of patients to low-intensity treatments compared with 78.5% of patients from the stepped care pathway who received low-intensity treatment first. Stratified care had significantly better depression treatment outcomes with 52.3% of the group reaching clinical recovery as measured by the PHQ-9 compared to 45.1% of patients in the stepped care group.
There were smaller differences in the treatment allocation of patients with anxiety disorders, potentially because clinical guidelines recommend many of these patients enter the model at step 3 anyway. It was noted that clinicians in the stepped care condition had low concordance with the stratified care algorithm.
Patients receiving stratified care had a higher median number of treatment sessions due to the evidence based guidance for step 3 treatments. Rates of patients dropping out of treatment was not significantly different between conditions. The mean additional cost for an episode of stratified care was just over £100.
The authors conclude:
The findings of this trial indicate that stratified care improves depression outcomes, albeit at an incremental cost per treatment.
A further and unexpected finding was that only patients defined as having ‘standard’ and not ‘complex’ presentations were more likely to reach clinical recovery in the stratified care condition. The paper debates the potential limitations of IAPT services for patients with complex and chronic presenting issues. There is also the consideration that stratified care is not as effective as predicted for ‘complex’ cases and that further research is needed to investigate improving treatment outcomes for this group of patients.
Strengths and limitations
This paper has given me much food for thought. There are issues that need some resolution in potential follow-up studies such as the predominance of a white and female sample, with relatively low mean age, considering the fact there is no upper age limit in IAPT services. We, therefore, cannot fully generalise these findings to the population as a whole and this study would benefit from replication with a focus on higher participants numbers from sections of the population who are older, younger, identifying as a gender other than female and from racial and ethnic groups other than white. Other factors may be important to consider in the algorithm such as physical health status, previous episodes of therapy or mental health care and supportive or protective factors in recovery. The step-up rate in the stratified care condition is almost 6% higher than the stepped care condition and the reasons for this would also be interesting to investigate.
Whilst I understand the ‘pragmatic approach’ taken in terms of not changing the content of treatment as usual, it is difficult to know if there were any differences in treating clinicians within or between the different participating services which may have impacted the results. Some form of standardisation of approach may have strengthened the outcome measurement – for example, two patients with similar presenting issues (for example generalised anxiety disorder) even within the same step and treatment modality may have received a different therapeutic model based on clinician experience or the training they have undertaken (for example Dugas or Borkovec).
Another issue not considered in any detail here, which is vital in clinical service provision, is the factor of patient choice of treatment modality. High-intensity interventions are not broken down in the analysis to their separate modalities of CBT, counselling etc. If patients with depression in the stratified care condition were offered a modality not in line with their own preference, such as CBT instead of counselling, this may also have had an impact on the outcome. In order to understand this more fully, it would seem logical to analyse results by step 3 modality rather than as a whole as ‘high-intensity interventions’ as some approaches may be more effective when using the algorithm than others.
That said, the methodology in this paper appears strong with good participant numbers and patient drop-out rates no higher than the expected level for IAPT services. In my view, there is much that could be taken from this study which could support decision making in clinical practice and enable a higher probability of clinical recovery for more patients presenting with depression. I think the algorithm shows promise and could be improved with further experimentation and perhaps focus purely on depression rather than including anxiety symptoms as well when these patients may already be starting their treatment episode at step 3.
Implications for practice
So what are the implications for clinical decision making? I keep coming back to the finding that decision making in the stepped care condition had low concordance with the algorithm. Should we be introducing a more standard and clinically reliable process in decision making regarding allocation to treatment modality? As a clinician who has always championed clinical judgement within the team, this paper has made me think long and hard about this.
Some IAPT services are starting to introduce AI decision making to their processes but in my opinion, this should only be after a clinician has completed a full individual assessment, including of risk, and should not be a replacement for information gathering or clinical judgement at the start of care. If an algorithm can be used to advise regarding allocation to treatment after this assessment, alongside clinical judgement and patient choice, and that allocation improves the probability of recovery, it can only be a good thing. It can never incorporate every nuance of every patient’s presentation and history and therefore should not replace a clinical assessment by a trained practitioner.
Finally, where does this leave the NICE Guidance consultation? There is no evidence here that removing stepped care as an approach is clinically required or financially effective. Within the stratified care condition which contains personalised treatment recommendations, 50.9% of patients were still allocated to low-intensity treatments first. The training and financial burden to services of offering high-intensity interventions to all patients as a first line treatment for depression makes this recommendation unfeasible and is likely to lead to longer waits for treatment for all patients and unnecessarily so. Our step 2 colleagues are experienced and skilled in what they do when the patients referred for low-intensity interventions are referred appropriately to them. In the same way that high-intensity interventions cannot support all patients to clinical recovery, neither can low intensity work and the key here (however, we make the decisions moving forwards) is to ensure that patients are not allocated to a treatment method which will not support their recovery given their presenting issues and history.
This study offers a very promising methodology for supporting clinicians to support patients more effectively and I am very interested to see where the authors may take this line of study next. With some refinements of the algorithm and further investigation of separate high-intensity modalities, I would certainly be one of the clinical leads volunteering to test this process out in service.
Statement of interests
Delgadillo J, Ali S, Fleck K, et al. (2022) Stratified Care vs Stepped Care for Depression: A Cluster Randomized Clinical Trial. JAMA Psychiatry. 2022;79(2):101-108. doi:10.1001/jamapsychiatry.2021.3539
BABCP (2022) Submission of BABCP response to NICE consultation – Depression in adults: treatment and management (update) https://babcp.com/NICEconsultJan2022
Kroenke, K., Spitzer, R. L., & Williams, J.B., (2001). The PHQ-9: Validity of a brief depression severity measure. Journal of General Internal Medicine, 16(9), 606–613.
National Collaborating Centre for Mental Health. The Improving Access to Psychological Therapies Manual, Version 4. Updated March 2020. Accessed February 27, 2022. https://www.england.nhs.uk/publication/the-improving-access-to-psychological-therapies-manual/
National Institute for Clinical Excellence (2021) Depression guideline for consultation https://nice.org.uk/guidance/gid-cgwave0725/documents/draft-guideline-4
Spitzer, R.L., Kroenke, K., Williams, J.B., Lowe, B., (2006) A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch Intern Med, 166: 1092-1097.