Research paperDevelopment of a prognostic model for predicting depression severity in adult primary patients with depressive symptoms using the diamond longitudinal study
Introduction
Mental health disorders account for 7.4% of the total disease burden with depression the main contributor (Whiteford et al., 2013). Most people seeking help for depressive symptoms are treated in primary care, (Australian Bureau of Statistics, 2011, Australian Institute of Health and Welfare, 2015) and around one quarter of primary care attendees report current depressive symptoms (Gunn et al., 2008; Herrman et al., 2002). Ten percent of attendees with subthreshold symptoms and no history of depression develop major depression over six months (Davidson et al., 2015) and 21% over two years (Karsten et al., 2011). Nearly 60% of those with current major depression meet criteria for major depression at least once over the next three years (Stegenga et al., 2012). Of primary care attendees satisfying criteria for major depression, around 50% are estimated to also have current anxiety (Gunn et al., 2008).
In a busy primary care practice, it can be difficult for clinicians to identify which patients with current depressive symptoms are likely to recover and which are likely to worsen, and to provide treatment appropriate for each trajectory. Primary care clinicians are often criticised for either over-treating patients with subthreshold depression (Davidson et al., 2015) or for not providing minimally adequate treatment for patients with major depression (Wang et al., 2007). One systematic approach to informing clinician's treatment decisions is to use a clinical prediction tool.
A clinical prediction tool is built around a prognostic model that uses clinical and psychosocial information to predict future depression severity. The clinical prediction tool uses the information provided by the prognostic model to stratify patients into different depression severity groups. Type and intensity of treatment is tailored to each group to optimise clinical outcomes with the least intensive treatment (Rubenstein et al., 2007). To date, no such clinical prediction tool exists that can be used to stratify primary care patients with depressive symptoms into different treatment options based on their predicted depressive symptoms.
We also conducted a literature search to identify existing prognostic models that would be suitable for inclusion in a clinical prediction tool that predicts future depressive symptoms in primary care patients with depressive symptoms, ranging from sub-threshold to severe. The literature search identified nine different prognostic models for depression developed using data from five unique primary care studies. Only two of the models focussed on predicting future depression within samples experiencing current depressive symptoms (Dowrick et al., 2011; Rubenstein et al., 2007). Of the remaining studies, three developed or validated prognostic models to predict the onset of depression (primary prevention) (Bellon et al., 2011; King et al., 2013; King et al., 2008), two studies developed a prediction rule to screen for the presence of current mood disorders (Vohringer et al., 2013; Zuithoff et al., 2009) and two studies developed algorithms to predict treatment response to antidepressants(Chekroud et al.; Perlis, 2013).
Of the two studies that developed prognostic models to predict future depression among people with current depressive symptoms, neither was suitable for inclusion in a clinical prediction tool (Dowrick et al., 2011; Rubenstein et al., 2007). In the first study, the prognostic model developed using trial data from 220 participants in the THREAD study was insufficiently robust to use in the clinical prediction tool because it had low prognostic accuracy (Dowrick et al., 2011). Furthermore, the development sample only included participants with mild to moderate depression, thus could not be generalised to new primary care patients who present with severe depression. The second study described the development of the Diagnostic Prognostic Index, which was derived using data from 1471 primary care attendees with current major depression participating in one of four randomised trials (Rubenstein et al., 2007). The Diagnostic Prognostic Index was also unsuitable because the development sample excluded patients with subthreshold depression. Given that in primary care subthreshold depression makes up the largest group of patients presenting with depressive symptoms, the prognostic model would not be generalisable to this population. Additionally, the Diagnostic Prognostic Index, consisting of over 60 items, would be too lengthy to administer in a primary care waiting room or during a consultation, limiting its usability and usefulness in routine clinical practice (Toll et al., 2008).
This study aimed to develop a prognostic model for future depression severity among adult primary care attendees with current depressive symptoms, ranging from sub-threshold to severe depression. To increase the utility and uptake of the clinical prediction tool we aimed to develop a model with relatively few items that were easy to collect in routine practice (Toll et al., 2008).
Section snippets
Source of data
We developed a prognostic model using data from the diamond (Diagnosis, Management and Outcomes of Depression) cohort study. Diamond is a 10-year prospective study of adult primary care patients with depressive symptoms (Gunn et al., 2008).
Cohort participants were recruited from 30 general practitioners (GPs) working at 30 different urban, regional and rural practices in Victoria, Australia between January and December 2005. Details of recruitment are published elsewhere (Gunn et al., 2008).
Participants
Distribution of participant characteristics in the development sample is shown in Table 1. The fraction of missing responses for each predictor variable for the development sample was small, ranging from zero for gender to 2.2% (13/593) for ever being afraid of a partner. Most participants (91%, 538/593) had complete data for the 20 candidate predictor variables, including the sensitive questions. Fourteen percent (82/593) had missing values for the outcome.
Model development
Table 2 shows the estimated
Discussion
We developed a brief, easily administered prognostic model to predict depression severity at three months in adult primary care patients with current depressive symptoms. The eight predictors were sex, depressive symptoms, current anxiety, history of depression, self-rated health, chronic physical illness, living alone, and perceived ability to manage available on income. The final model consists of 17 questions, nine of which are from the PHQ-9. Including potentially sensitive or distressing
Conclusion
We developed a brief, easily administered prognostic model for use in primary care across the depressive symptom range to predict depression severity at three months. A clinical prediction tool utilising this model has the potential to assist clinicians manage the large burden of mental health symptoms presenting to primary care. Widespread implementation of tools like this offers the best chance of ensuring that limited resources are allocated based on need.
Acknowledgements
We acknowledge the 30 dedicated general practitioners, their patients, and practice staff for making this research possible. We thank the cohort participants for their ongoing involvement in the study. We also thank the diamond project team and associate investigators involved in the study: A/Prof. Lena Sanci, A/Prof Cathy Mihalopoulos, Ms Maria Potiriadis, Ms Konstancja Densley, Ms Aves Middleton, and the casual research staff.
The authors submit this manuscript on behalf of the diamond study
Contributors
PC, SD and JG drafted the manuscript. PC conducted the analyses and produced the tables and figures. SD, GG, CD, FG, KH, HH, JG, and PC formed the multi-disciplinary expert group to identify and select candidate predictor variables. RW provided statistical expertise on the development of the prognostic model. All the authors contributed to development and drafting of the manuscript
Funding sources
The diamond study was initiated with pilot funding from the Victoria Centre of Excellence in Depression, Anxiety
References (33)
- et al.
Mental health interventions and future major depression among primary care patients with subthreshold depression
J. Affect. Disord.
(2015) - et al.
A trajectory-based approach to understand the factors associated with persistent depressive symptoms in primary care
J. Affect. Disord.
(2013) - et al.
Psychiatry and primary care: the patient health questionnaire somatic, anxiety, and depressive symptom scales: a systematic review
Gen. Hosp. Psychiatry
(2010) A clinical risk stratification tool for predicting treatment resistance in major depressive disorder
Biol. Psychiatry
(2013)- et al.
Validation, updating and impact of clinical prediction rules: a review
J. Clin. Epidemiol.
(2008) - et al.
A clinical predictive score for mood disorder risk in low-income primary care settings
J. Affect. Disord.
(2013) - et al.
Use of mental health services for anxiety, mood, and substance disorders in 17 countries in the WHO world mental health surveys
Lancet
(2007) - et al.
Global burden of disease attributable to mental and substance use disorders: findings from the Global Burden of Disease Study 2010
Lancet
(2013) - et al.
Prognosis and prognostic research: validating a prognostic model
BMJ
(2009) - Australian Bureau of Statistics, 2011. 4329.0 - Characteristics of people using mental health services and prescription...
Predicting the onset of major depression in primary care: international validation of a risk prediction algorithm from Spain
Psychol. Med.
Cross-trial prediction of treatment outcome in depression: a machine learning approach
Lancet Psychiatry Publ. Online Jan.
Estimating probability of sustained recovery from mild to moderate depression in primary care: evidence from the THREAD study
Psychol. Med.
Target-D: a stratified individually randomized controlled trial of the diamond clinical prediction tool to triage and target treatment for depressive symptoms in general practice: study protocol for a randomized controlled trial
Trials.
Who is identified when screening for depression is undertaken in general practice? Baseline findings from the Diagnosis, Management and Outcomes of Depression in Primary Care (diamond) longitudinal study
Med. J. Aust.
Cited by (13)
Predicting clinically significant response to primary care treatment for depression from electronic health records of veterans
2021, Journal of Affective DisordersCitation Excerpt :Clinical prediction tools can guide clinical decision-making by stratifying patients into different risk categories using statistical models that use large data sources. Although several such tools predict trajectories of depression in currently depressed patients (Rubenstein et al., 2007, Dowrick et al., 2011, Chondros et al., 2018), prior studies have not examined the likelihood of response to treatment in primary care settings that include integrated care. These studies have also not typically used data from electronic health records (EHR), but rather from longitudinal cohorts.
Clinical efficacy of a Decision Support Tool (Link-me) to guide intensity of mental health care in primary practice: a pragmatic stratified randomised controlled trial
2021, The Lancet PsychiatryCitation Excerpt :The second component is a recommendation for treatment pathways (low-intensity or high-intensity care), which depend on the predicted symptom severity. Link-me is based on our previously developed Target-D approach that focused on depression and used the predicted severity of a patient's depressive symptoms at 3 months to match them to either an online programme (low intensity care) or to nurse-led collaborative care or so-called care navigation (high-intensity care).13–15 In Link-me, we refined the approach by updating the prognostic model for depression and adding one for anxiety, providing a broader range of low-intensity treatment options, expanding the care navigation workforce to include other health professionals (ie, health professionals who are not nurses or trained mental health specialists), and reducing the financial barriers to care for care navigation recipients.
Developing prediction models for total knee replacement surgery in patients with osteoarthritis: Statistical analysis plan
2020, Osteoarthritis and Cartilage OpenLink-me: Protocol for a randomised controlled trial of a systematic approach to stepped mental health care in primary care
2019, Contemporary Clinical TrialsCitation Excerpt :In this context, three PHNs have been selected to work with the University of Melbourne on Link-me, a randomised controlled trial (RCT) of a systematic approach to stepped mental health care. Link-me builds on over 15 years of research, including the development of a clinical prediction tool (CPT) from a ten year longitudinal cohort study of depression and preliminary testing of that tool in a systematic approach to identification of and tailored treatment for depression in the Target-D RCT [14–16]. Link-me uses a Decision Support Tool (DST) completed by patients in the general practice setting to predict the future severity of depressive and anxiety symptoms, and triage patients into high or low intensity care accordingly.
Teacher depression as a dynamic variable: Exploring the nature and predictors of change over the head start year
2019, Journal of Applied Developmental PsychologyCitation Excerpt :Depression is rooted in a complex array of chemical reactions in the brain, which are in turn affected by a large constellation of factors, including those of biological (e.g., genetics, hormones, illness), environmental (e.g., diet, stressful or traumatic events) and cognitive (e.g., coping skills) provenance (Gotlib & Hammen, 2014; Rantala, Luoto, Krams, & Karlsson, 2018). Accordingly, depression is a dynamic condition, with some suffers experiencing temporary bouts, while other find that symptoms wax and wane over the course of their lifetime (Chondros et al., 2018; Heinz et al., 2018). At any given time, though, the prevalence of depression is substantially higher – up to 1.7 times greater – among women (Albert, 2015).