Predicting post-treatment symptom severity for adults receiving psychological therapy in routine care for generalised anxiety disorder: a machine learning approach

Approximately half of generalised anxiety disorder (GAD) patients do not recover from first-line treatments, and no validated prediction models exist to inform individuals or clinicians of potential treatment benefits. This study aimed to develop and validate an accurate and explainable prediction model of post-treatment GAD symptom severity. Data from adults receiving treatment for GAD in eight Improving Access to Psychological Therapies (IAPT) services (n=15,859) were separated into training, validation and holdout datasets. Thirteen machine learning algorithms were compared using 10-fold cross-validation, against two simple clinically relevant comparison models. The best-performing model was tested on the holdout dataset and model-specific explainability measures identified the most important predictors. A Bayesian Additive Regression Trees model out-performed all comparison models (MSE=16.54 [95 % CI=15.58; 17.51]; MAE=3.19; R²=0.33, including a single predictor linear regression model: MSE=20.70 [95 % CI=19.58; 21.82]; MAE=3.94; R²=0.14). The five most important predictors were: PHQ-9 anhedonia, GAD-7 annoyance/irritability, restlessness and fear items, then the referral-assessment waiting time. The best-performing model accurately predicted post-treatment GAD symptom severity using only pre-treatment data, outperforming comparison models that approximated clinical judgement and remaining within the GAD-7 error of measurement and minimal clinically important differences. This model could inform treatment decision-making and provide desired information to clinicians and patients receiving treatment for GAD.

Examining bi-directional change in sleep and depression symptoms in individuals receiving routine psychological treatment

BACKGROUND

Sleep disturbance is a common symptom of depression. There is conflicting evidence whether improvements in sleep might impact depressive symptoms, or whether treating the core depressive symptoms might improve sleep disturbance. This study explored the bi-directional impact of sleep and depressive symptom change among individuals receiving psychological treatment.

METHODS

Session-by-session change in sleep disturbance and depressive symptom severity scores were explored in patients receiving psychological therapy for depression from Improving Access to Psychological Therapies services in England. Bi-directional change in sleep disturbance and depressive symptoms was modelled using random-intercept cross-lagged panel models with items from the PHQ-9.

RESULTS

The sample included 17,732 adults that had received three or more treatment sessions. Both depressive symptoms and sleep disturbance scores decreased. Between initial timepoints, higher sleep disturbance was associated with lower depression scores, but after this point positive cross-lagged effects were observed for both the impact of sleep disturbance on later depressive symptoms, and depressive symptoms on later sleep disturbance scores. The magnitude of effects suggested depressive symptoms may have more impact on sleep than the reverse, and this effect was larger in sensitivity analyses.

CONCLUSIONS

Findings provide evidence that psychological therapy for depression results in improvements in core depressive symptoms and sleep disturbance. There was some evidence that depressive symptoms may have more impact on sleep disturbance scores at the next therapy session, than sleep disturbance does on later depressive symptoms. Targeting the core symptoms of depression initially may optimise outcomes, but further research is needed to elucidate these relationships.

Predicting prognosis for adults with depression using individual symptom data: a comparison of modelling approaches

BACKGROUND

This study aimed to develop, validate and compare the performance of models predicting post-treatment outcomes for depressed adults based on pre-treatment data.

METHODS

Individual patient data from all six eligible randomised controlled trials were used to develop (k = 3, n = 1722) and test (k = 3, n = 918) nine models. Predictors included depressive and anxiety symptoms, social support, life events and alcohol use. Weighted sum scores were developed using coefficient weights derived from network centrality statistics (models 1-3) and factor loadings from a confirmatory factor analysis (model 4). Unweighted sum score models were tested using elastic net regularised (ENR) and ordinary least squares (OLS) regression (models 5 and 6). Individual items were then included in ENR and OLS (models 7 and 8). All models were compared to one another and to a null model (mean post-baseline Beck Depression Inventory Second Edition (BDI-II) score in the training data: model 9). Primary outcome: BDI-II scores at 3-4 months.

RESULTS

Models 1-7 all outperformed the null model and model 8. Model performance was very similar across models 1-6, meaning that differential weights applied to the baseline sum scores had little impact.

CONCLUSIONS

Any of the modelling techniques (models 1-7) could be used to inform prognostic predictions for depressed adults with differences in the proportions of patients reaching remission based on the predicted severity of depressive symptoms post-treatment. However, the majority of variance in prognosis remained unexplained. It may be necessary to include a broader range of biopsychosocial variables to better adjudicate between competing models, and to derive models with greater clinical utility for treatment-seeking adults with depression.

Role of age, gender and marital status in prognosis for adults with depression: An individual patient data meta-analysis

AIMS

To determine whether age, gender and marital status are associated with prognosis for adults with depression who sought treatment in primary care.

METHODS

Medline, Embase, PsycINFO and Cochrane Central were searched from inception to 1st December 2020 for randomised controlled trials (RCTs) of adults seeking treatment for depression from their general practitioners, that used the Revised Clinical Interview Schedule so that there was uniformity in the measurement of clinical prognostic factors, and that reported on age, gender and marital status. Individual participant data were gathered from all nine eligible RCTs (N = 4864). Two-stage random-effects meta-analyses were conducted to ascertain the independent association between: (i) age, (ii) gender and (iii) marital status, and depressive symptoms at 3-4, 6-8, and 9-12 months post-baseline and remission at 3-4 months. Risk of bias was evaluated using QUIPS and quality was assessed using GRADE. PROSPERO registration: CRD42019129512. Pre-registered protocol https://osf.io/e5zup/.

RESULTS

There was no evidence of an association between age and prognosis before or after adjusting for depressive 'disorder characteristics' that are associated with prognosis (symptom severity, durations of depression and anxiety, comorbid panic disorderand a history of antidepressant treatment). Difference in mean depressive symptom score at 3-4 months post-baseline per-5-year increase in age = 0(95% CI: -0.02 to 0.02). There was no evidence for a difference in prognoses for men and women at 3-4 months or 9-12 months post-baseline, but men had worse prognoses at 6-8 months (percentage difference in depressive symptoms for men compared to women: 15.08% (95% CI: 4.82 to 26.35)). However, this was largely driven by a single study that contributed data at 6-8 months and not the other time points. Further, there was little evidence for an association after adjusting for depressive 'disorder characteristics' and employment status (12.23% (-1.69 to 28.12)). Participants that were either single (percentage difference in depressive symptoms for single participants: 9.25% (95% CI: 2.78 to 16.13) or no longer married (8.02% (95% CI: 1.31 to 15.18)) had worse prognoses than those that were married, even after adjusting for depressive 'disorder characteristics' and all available confounders.

CONCLUSION

Clinicians and researchers will continue to routinely record age and gender, but despite their importance for incidence and prevalence of depression, they appear to offer little information regarding prognosis. Patients that are single or no longer married may be expected to have slightly worse prognoses than those that are married. Ensuring this is recorded routinely alongside depressive 'disorder characteristics' in clinic may be important.

The importance of transdiagnostic symptom level assessment to understanding prognosis for depressed adults: analysis of data from six randomised control trials

BACKGROUND

Depression is commonly perceived as a single underlying disease with a number of potential treatment options. However, patients with major depression differ dramatically in their symptom presentation and comorbidities, e.g. with anxiety disorders. There are also large variations in treatment outcomes and associations of some anxiety comorbidities with poorer prognoses, but limited understanding as to why, and little information to inform the clinical management of depression. There is a need to improve our understanding of depression, incorporating anxiety comorbidity, and consider the association of a wide range of symptoms with treatment outcomes.

METHOD

Individual patient data from six RCTs of depressed patients (total n = 2858) were used to estimate the differential impact symptoms have on outcomes at three post intervention time points using individual items and sum scores. Symptom networks (graphical Gaussian model) were estimated to explore the functional relations among symptoms of depression and anxiety and compare networks for treatment remitters and those with persistent symptoms to identify potential prognostic indicators.

RESULTS

Item-level prediction performed similarly to sum scores when predicting outcomes at 3 to 4 months and 6 to 8 months, but outperformed sum scores for 9 to 12 months. Pessimism emerged as the most important predictive symptom (relative to all other symptoms), across these time points. In the network structure at study entry, symptoms clustered into physical symptoms, cognitive symptoms, and anxiety symptoms. Sadness, pessimism, and indecision acted as bridges between communities, with sadness and failure/worthlessness being the most central (i.e. interconnected) symptoms. Connectivity of networks at study entry did not differ for future remitters vs. those with persistent symptoms.

CONCLUSION

The relative importance of specific symptoms in association with outcomes and the interactions within the network highlight the value of transdiagnostic assessment and formulation of symptoms to both treatment and prognosis. We discuss the potential for complementary statistical approaches to improve our understanding of psychopathology.

Application of audio fingerprinting to neonatal EEG

A clinical neurophysiologist must recognize patterns in EEG signals to evaluate the health of a patient's brain activity. Rare or unusual patterns may take time to correctly identify. The ability to automatically assist this recall would be beneficial in ensuring that appropriate measures could be taken in a timely fashion. Audio fingerprinting is a method used to identify songs using only a snippet of the song. Fingerprints are extracted from a sub-section of the song and matched against a database of previously computed fingerprints. In this paper, a fingerprint quantization technique is implemented on neonatal EEG data to attempt to identify sections of EEG data when only seeing a sub-section of the data. The impact of signal distortions is investigated and results from a database of one hour recordings from 40 newborns are presented.

Stream computing for biomedical signal processing: A QRS complex detection case-study

Recent developments in "Big Data" have brought significant gains in the ability to process large amounts of data on commodity server hardware. Stream computing is a relatively new paradigm in this area, addressing the need to process data in real time with very low latency. While this approach has been developed for dealing with large scale data from the world of business, security and finance, there is a natural overlap with clinical needs for physiological signal processing. In this work we present a case study of streams processing applied to a typical physiological signal processing problem: QRS detection from ECG data.

Molecular epidemiology of Mycobacterium abscessus complex isolates in Ireland

BACKGROUND

The Mycobacterium abscessus complex are the rapidly growing mycobacteria (RGM) most commonly causing lung disease, especially in cystic fibrosis (CF) patients. Ireland has the world's highest CF incidence. The molecular epidemiology of M. abscessus complex in Ireland is unreported.

METHODS

We performed rpoB gene sequencing and multi-locus sequence typing (MLST) on M. abscessus complex strains isolated from thirty-six patients in 2006-2012 (eighteen known CF patients).

RESULTS

Twenty-eight strains (78%) were M. abscessus subsp. abscessus, eight M. abscessus subsp. massiliense, none were M. abscessus subsp. bolletii. Sequence type 1 (ST1) and ST26 (M. abscessus subsp. abscessus) were commonest. Seven M. abscessus subsp. abscessus STs (25%) were novel (two with novel alleles). Seven M. abscessus subsp. massiliense STs were previously reported (88%), including two ST23, the globally successful clone. In 2012, of 552 CF patients screened, eleven were infected with M. abscessus complex strains (2%).

CONCLUSIONS

The most prevalent M. abscessus subsp. abscessus and M. abscessus subsp. massiliense strains in Ireland belong to widely-distributed STs, but there is evidence of high M. abscessus subsp. abscessus diversity.

Meeting breast cancer patients' information needs during radiotherapy: what can we do to improve the information and support that is currently provided?

Previous research has reported that patients require specific information relating to radiotherapy; however, these studies fail to describe patients' specific information needs over time. The aims of this study were to determine the specific information needs of breast cancer patients who are receiving radiotherapy and identify when patients prefer to receive specific information. Semi-structured interviews were conducted with 34 early breast cancer patients and 14 health professionals. Seventeen patients were interviewed after treatment completion, and 17 patients were interviewed on at least two occasions during their radiotherapy. Grounded theory and the constant comparative method were used to analyse the data. Three main categories emerged from the data: 'repertoire of information', 'amount of information relating specifically to radiotherapy' and'tailoring information to match patients' radiotherapy journeys'. Patients' information needs were identified, and key messages and strategies to inform patients were described. This paper identifies breast cancer patient's specific information needs during radiotherapy and shows that patients' information needs are highest during their first appointment with their radiation oncologist and at the time of their planning appointment. The findings presented will enable health professionals to develop and refine their approaches to patient education in radiotherapy.

Large-scale molecular analysis of a 34 Mb interval on chromosome 6q: major refinement of the RP25 interval

A large scale bioinformatics and molecular analysis of a 34 Mb interval on chromosome 6q12 was undertaken as part of our ongoing study to identify the gene responsible for an autosomal recessive retinitis pigmentosa (arRP) locus, RP25. Extensive bioinformatics analysis indicated in excess of 110 genes within the region and we also noted unfinished sequence on chromosome 6q in the Human Genome Database, between 58 and 61.2 Mb. Forty three genes within the RP25 interval were considered as good candidates for mutation screening. Direct sequence analysis of the selected genes in 7 Spanish families with arRP revealed a total of 244 sequence variants, of which 67 were novel but none were pathogenic. This, together with previous reports, excludes 60 genes within the interval ( approximately 55%) as disease causing for RP. To investigate if copy number variation (CNV) exists within RP25, a comparative genomic hybridization (CGH) analysis was performed on a consanguineous family. A clone from the tiling path array, chr6tp-19C7, spanning approximately 100-Kb was found to be deleted in all affected members of the family, leading to a major refinement of the interval. This will eventually have a significant impact on cloning of the RP25 gene.

Linkage validation of RP25 Using the 10K genechip array and further refinement of the locus by new linked families

Retinitis pigmentosa (RP) is a clinically and genetically heterogeneous group of retinal dystrophies, characterised by rod photoreceptor cell degeneration with autosomal recessive RP (arRP) as the commonest form worldwide. To date, a total of 26 loci have been reported for arRP, each having a prevalence of 1-5%, except for the RP25 locus which was identified as the genetic cause of 14% of arRP cases in Spain. In order to validate the original linkage of RP25, we undertook a total genome scan using the 10K GeneChip mapping array on three of the previously linked families. The data obtained supported the initial findings of linkage. Additionally, linkage analysis in 18 newly ascertained arRP families was performed using microsatellite markers spanning the chromosome 6p12.1-q15 interval. Five out of the 18 families showed suggestive evidence of linkage to RP25, hence supporting the high prevalence of this locus in the Spanish population. Furthermore, the finding of a crossover in one of these families is likely to have refined the disease interval from the original 16 cM to only a 2.67 cM region between D6S257 and D6S1557.

Carbonylation of glycolytic proteins is a key response to drug-induced oxidative stress and apoptosis

Recent work has highlighted the importance of protein post-translational modifications such as phosphorylation (enzymatic) and nitrosylation (nonenzymatic) in the early stages of apoptosis. In this study, we have investigated the levels of protein carbonylation, a nonenzymatic protein modification that occurs in conditions of cellular oxidative stress, during etopside-induced apoptosis of HL60 cells. Within 1 h of VP16 treatment, a number of proteins underwent carbonylation due to oxidative stress. This was inhibited by the antioxidant N-acetyl-L-cysteine. Among the proteins found to be carbonylated were glycolytic enzymes. Subsequently, we found that the rate of glycolysis was significantly reduced, probably due to a carbonylation mediated reduction in enzymatic activity of glycolytic enzymes. Our work demonstrates that protein carbonylation can be rapidly induced through cytotoxic drug treatment and may specifically inhibit the glycolytic pathway. Given the importance of glycolysis as a source of cellular ATP, this has severe implications for cell function.

Mucus as a barrier to the permeability of hydrophilic and lipophilic compounds in the absence and presence of sodium taurocholate micellar systems using cell culture models

A mucus secreting CaCo-2-Ht29GlucH cell co-culture model was characterised and used to examine the influence of mucus as a barrier to the transport of hydrophilic and lipophilic compounds in the absence and presence of sodium taurocholate micellar (NaTC) systems. TEER measurements and permeability studies using the hydrophilic markers (mannitol, polyethylene glycols (PEGS) 900 and 4000) indicated that the paracellular permeability of the co-culture model was greater than that of the CaCo-2 model. At pH 7.4, no difference in the transport of a model lipophilic drug, dextropropoxyphene, was observed between the two models. However, at pH 4.5, when the drug was highly ionised the transport was significantly lower across the co-culture monolayers. NaTC micellar systems appeared to affect the different cell culture models in the order CaCo-2>CaCo-2-Ht29GlucH>Ht29GlucH. Following removal of the mucus layer by incubation with the mucolytic agent, N-acetyl-l-cysteine, the absorption enhancing potential of NaTC micellar systems was increased in the co-culture model.

The TAPS Project. 10: The long-stay populations of Friern and Claybury hospitals. The baseline survey

As part of the TAPS project, the baseline long-stay populations of Friern and Claybury Hospitals were compared using the TAPS assessment schedules. Each hospital has reduced bed numbers by two-thirds, and although the two have been subjected to different influences, their remaining long-stay patients were similar, and characterised by severe problems with hygiene, florid psychotic symptoms, a high prevalence of negative symptoms, and impoverished social networks.

The TAPS Project. 7: Mental hospital closure--a literature review of outcome studies and evaluative techniques

Various techniques have been used to assess the results of 'deinstitutionalisation'. This paper reviews these outcome studies, most of which have concluded that a new long-stay group will replace the old long-stay patients, and that a residual group of ageing patients, mostly with organic syndromes, is difficult to accommodate in the community.

The TAPS Project. 13: Clinical and social outcomes of long-stay psychiatric patients after one year in the community

The first prospective, controlled study of a large, long-stay, in-patient population as their hospital services were reprovided in the community is reported. Two-hundred-and-seventy-eight patients were matched individually with similar patients who remained in hospital. Clinical and social outcomes over a one-year follow-up are presented for the first three years of the reprovision process. Compared with matches, the leavers at follow-up had more diverse social networks that contained a higher proportion of contacts named as friends. The care facilities in the community allowed more opportunities for patient autonomy than the hospitals across all aspects of the environments measured. At follow-up, more leavers than matches wanted to remain in their current placement, while fewer leavers than matches said there was nothing they liked about their current placement. More leavers than matches found their medication helpful.

The TAPS Project. 9: The reliability of the Patient Attitude Questionnaire

The development of the hospital and community versions of the Patient Attitude Questionnaire is described. The instrument rates the attitudes of psychiatric patients towards their treatment settings and staff, and is framed specifically to assess attitudinal change during the transfer of patients from hospital. For the items rated using the kappa coefficient of agreement, the mean test-retest reliability value was 0.51, and the average inter-rater value was 0.82. This study shows that long-term psychiatric patients are able to give clear and consistent views about their living arrangements--views that should be sought and respected by staff.

The TAPS Project. 5: The structure of social-network data obtained from long-stay patients

The Social Network Schedule was successfully completed on 489 (64%) of 770 long-stay patients in Friern and Claybury Hospitals and the data on 3774 social contacts of the patients subjected to log-linear modelling. This showed that the category 'confidant' was salient in understanding the data. Latent-class analysis highlighted the importance of the nature of the social interaction between patients and their contacts. In particular, non-verbal exchanges of goods and services emerged as important in creating social bonds.

The TAPS Project. 4: An observational study of the social life of long-stay patients

Observations of long-stay patients in a club in a psychiatric hospital were carried out unobtrusively over several months. Much of the social activity that went on was stimulated by an economy based around the purchase and exchange of tea and cigarettes, and the borrowing and lending of small change. Most patients could be fitted into one of four social categories: helpers, friends, nuisances, and the asocial. The inter-rater reliability of the observations was found to be 0.95. A close correspondence emerged between the observations of patients' social behaviour and material reported by patients when interviewed with the Social Network Schedule.

Chronically ill psychiatric patients in a district general hospital unit. A survey and two-year follow-up in an inner-London health district

A survey was undertaken at the end of 1984 of all patients occupying psychiatric beds for more than six months in an inner-London health district. Excluding those with senile dementia, 30 patients were identified. Two years later, a follow-up survey traced the original cohort and the accumulation of additional patients on the wards. The finding that the rate of discharge from the original group matches the accumulation of new patients into the survey suggests a need for new types of community provision within the district.

The TAPS project. 3: Predicting the community costs of closing psychiatric hospitals

The planning of long-term care in the community as an alternative to in-patient care requires accurate information on the likely expense of altering the balance of provision. Unfortunately, as very few long-stay psychiatric hospitals have yet closed, the planning of these resource requirements has had to proceed in a vacuum. By examining the costs of community reprovision for the first 136 people to leave Claybury and Friern Hospitals, a prediction equation has been estimated from existing data which links the hospital-assessed characteristics (including psychiatric symptoms and behavioural problems) of these people to the subsequent cost of community care. About a third of the observed variation in these costs can be explained statistically by these 'baseline' characteristics. However, the first cohorts exhibit fewer behavioural problems and other symptoms of mental illness, they have been in hospital for shorter lengths of time, and they are younger. The prediction equation for the leavers is thus used to extrapolate community costs for those hospital residents yet to leave. It is found that community costs are lower than hospital costs, not just for the first cohorts of leavers, but for the full populations of the two hospitals scheduled to close.