The British Journal of Psychiatry (2017)

210, 4753. doi: 10.1192/bjp.bp.116.189027

Impact of long-term medical conditions

on the outcomes of psychological therapy
for depression and anxiety
Jaime Delgadillo, Alexander Dawson, Simon Gilbody and Jan R. Bohnke

Background
Long-term conditions often coexist with depression and
anxiety.
Aims
To assess the effectiveness of stepped-care psychological
therapies for patients with long-term conditions.
Method
Data from 28 498 patients were analysed using regression to
model depression (Patient Health Questionnaire (PHQ-9)) and
anxiety (Generalised Anxiety Disorder scale (GAD-7))
outcomes. Post-treatment symptoms and effect sizes (d)
were estimated for individuals with and without long-term
conditions, controlling for covariates. The likelihood of access
and response to intensive psychological interventions was
also examined.
Results
Higher post-treatment symptoms were predicted for
patients with musculoskeletal problems (d = 0.220.27),
chronic obstructive pulmonary disease (d = 0.260.33),
diabetes (d = 0.050.13) and psychotic disorders (d = 0.50
0.58). Most long-term conditions were associated with
greater odds of accessing high-intensity therapies, yet
individuals who accessed these continued to have higher
average post-treatment symptoms.
Conclusions
Some long-term conditions are associated with greater
intensity of care and poorer outcomes after therapy.
Declaration of interest
None.
Copyright and usage
B The Royal College of Psychiatrists 2017.

Long-term conditions are highly prevalent in the general population

Method
affecting approximately 20% of people1 above the age of 16 and 58%
of people over the age of 60.2 Among the more prevalent long-term Setting and interventions
conditions in the general population are hypertension, chronic pain,
Retrospective clinical case records for a cohort of patients receiving
gastrointestinal disorders, asthma, arthritis, diabetes, heart disease
psychological therapy were analysed. These data were gathered as
and chronic obstructive pulmonary disease (COPD).1,3 Multi-
part of routine clinical care at a psychological therapy service in
morbidity is common3 and these long-term conditions carry a huge
the north of England that was linked to the national Improving
financial cost to health services, accounting for approximately 70%
Access to Psychological Therapies (IAPT) programme.7 The study
of total health and social care expenditure in public healthcare
was conducted as a service evaluation using fully de-identified
systems like the UK National Health Service.2 This highlights
data, and therefore did not require ethical approval. Patients
the need to treat both long-term conditions and mental health
provided informed consent for their anonymous data to be used
problems concurrently. Although access to psychological care
for audit, evaluation and research purposes. Patients accessing
has been recommended as part of integrated care for people with
the service presented with depression- and anxiety-related
long-term conditions,4 it is possible that the effect of psychological
problems and received evidence-based interventions organised
interventions for mental health problems may be attenuated by
in a stepped-care model.8 The majority accessed brief (less than
long-term conditions. For example, Dickens and collaborators5
eight sessions), low-intensity guided self-help interventions based
reviewed the efficacy of psychological interventions for depression
on principles of CBT. Those with enduring symptoms after
in patients with coronary heart disease and concluded that clinical
guided self-help and those with more severe conditions had
trials show mixed evidence with small effect sizes favouring
access to high-intensity (up to 20 sessions) therapies including
cognitivebehavioural therapy (CBT) as the treatment of choice
CBT, interpersonal psychotherapy, counselling for depression
for this population. Similar conclusions were reached in a
and eye movement desensitisation and reprocessing (EMDR) for
systematic review of psychological interventions for persistent
post-traumatic stress.
pain, which indicates modest effects for CBT in depression and
anxiety symptom reductions.6 Clinical trials reviewed in these
meta-analyses often include participants with highly disabling
long-term conditions treated in specialist services or hospital Measures and data sources
settings, and therefore it is unclear whether these findings are Consistent with the national IAPT programme, patients in the
reflective of outcomes from routinely delivered therapy in primary sample were asked to self-complete three standard outcome
care settings, where the majority of patients long-term conditions measures to monitor progress on a weekly basis. The Patient
are treated. The objective of this study was to investigate the Health Questionnaire (PHQ-9) is a nine-item measure of major
clinical effectiveness of primary care psychological interventions depression symptoms.9 Each item is rated on four ordinal
for depression and anxiety in a large naturalistic sample, comparing response options (0, not at all; 3, nearly every day), resulting in
outcomes between patients with and without long-term conditions. a depression severity score between 0 and 27. A cut-off 510 is

47
Delgadillo et al

used to detect clinically significant depression symptoms. The and 36.6% were unemployed. Primary presenting problems
Generalised Anxiety Disorder scale (GAD-7) is a seven-item recorded in clinical assessments are presented in Table 1; the most
measure of anxiety symptoms; it is also rated using four ordinal common were recurrent depression (38.5%), mixed anxiety and
response options (0, not at all; 3, nearly every day), resulting depression (26.0%) and generalised anxiety disorder (12.1%).
in an anxiety severity score between 0 and 21.10 A cut-off Self-reported long-term conditions are listed in order of
score 58 is recommended to identify the likely presence of a prevalence in Table 2. Overall, 23.2% of patients reported having
diagnosable anxiety disorder. The Work and Social Adjustment at least one long-term condition; the most common were asthma
Scale (WSAS) is a measure of functioning across five domains: (6.8%), musculoskeletal problems (chronic pain, 1.8%) and
work, home management, social leisure activities, private leisure hypertension (1.7%). Approximately 68.0% of patients in this
activities, and family and relationships.11 Each item is rated sample only received low-intensity guided self-help interventions,
between 0 (no impairment) and 8 (very severe impairment), with and 32.0% accessed high-intensity therapies.
a total severity score between 0 and 40.
De-identified clinical assessment records were also collected Statistical analysis
for participants, including demographic (age, gender, ethnicity,
employment, socioeconomic deprivation) and clinical information The goals of the analysis were to predict depression (PHQ-9)
(primary diagnosis, baseline severity, use of psychotropic and anxiety (GAD-7) symptom severity at the end of treatment,
medication, number of treatment contacts, number of referrals controlling for the demographic and clinical characteristics
into the service for each patient over a 5-year period, pre- and described above, and to compare these outcomes between
post-treatment outcome measures described above). Self-reported individuals with and without long-term conditions. Given the
long-term conditions were gathered using a standardised checklist typically high correlations between the PHQ-9 and GAD-7
of chronic illnesses at the time of initial assessment.12 This scales,14 we applied a seemingly unrelated regression (SUR)
checklist prompts clinicians to gather information about 15 model.15 SUR models estimate several equations simultaneously
specific conditions including severe (psychotic) mental health when the error terms of these equations are potentially correlated,
problems and an option to note other unspecified conditions. which is a likely scenario for the prediction equations from
Socioeconomic deprivation was assessed by matching participants PHQ-9 and GAD-7 scores. These models have been applied
home postcodes to the English Index of Multiple Deprivation previously to model correlated dimensions of patient-reported
(IMD)13 and categorising participants into quintile levels of outcomes and have been shown to increase the efficiency of
deprivation (categorical variable named IMD). estimates in such situations.16 As a sensitivity analysis, we repeated
the SUR analysis using a dichotomous variable denoting the
presence or absence of a long-term condition, instead of entering
Sample characteristics dummy variables for the different long-term condition categories.
A total of 32 734 case records were gathered for patients who In order to compare outcomes between patients with and without
accessed psychological treatment (at least one session) over a long-term conditions, we estimated effect sizes (Cohens d) based
period of 5 years between 2010 and 2015. Of these, 2676 (8.2%) on comparing post-treatment outcomes between specific long-
were excluded because they had no recorded information about term condition groups v. the no long-term conditions category
long-term conditions, and a further 1560 (4.8%) were excluded and the root mean square error (RMSE) of the respective part
because no baseline and end scores were available for at least (i.e. PHQ-9 or GAD-7) of the SUR model.
one of the outcome measures (PHQ-9 or GAD-7). This resulted As a sensitivity analysis, we assessed differential item functioning
in a sample of 28 498 case records that were available for analysis. (DIF) to verify whether responses to the PHQ-9 and GAD-7
The mean age in the sample was 38.27 (s.d. = 13.94, range 1692); measures are comparable across long-term condition groups, or
64.6% were female; 85.5% were of a White British background; whether differences in outcome scores could be biased in favour,
or against, certain long-term conditions. Specifically, we aimed
to assess whether responses to individual PHQ-9 or GAD-7 items
corresponded to the same psychopathology severity levels across
Table 1 Primary presenting problems recorded in clinical long-term condition groups. To achieve this, we applied logistic
assessments ( n = 28 498)
Description n %
Table 2 Self-reported long-term medical conditions ( n = 28 498)
Recorded presenting problem (n = 22 609)
Description n %
Recurrent depression 8698 38.5
Mixed anxiety and depression 5887 26.0 None 21 882 76.8
Generalised anxiety disorder 2725 12.1 Other (unspecified) 2316 8.1
Depressive episode 1160 5.1 Asthma 1935 6.8
Panic disorder 970 4.3 Chronic musculoskeletal 507 1.8
Social phobia 540 2.4 Hypertension 490 1.7
Obsessivecompulsive disorder 643 2.8 Non-insulin-dependent diabetes mellitus 327 1.1
Post-traumatic stress disorder 565 2.5 Epilepsy 188 0.7
Specific phobia 290 1.3 Chronic obstructive pulmonary disease 137 0.5
Bereavement 201 0.9 Coronary heart disease 137 0.5
Eating disorder 177 0.8 Severe (psychotic) mental health problems 134 0.5
Agoraphobia 163 0.7 Insulin-dependent diabetes mellitus 111 0.4
Somatoform disorder 149 0.7 Cancer 114 0.4
Alcohol related mental or behavioural disorder 40 0.2 Stroke and transient ischaemic attack 76 0.3
Bipolar affective disorder 22 0.1 Multiple sclerosis 49 0.2
Does not meet diagnostic criteria for a Chronic kidney disease 41 0.1
common mental disorder 379 1.7 Heart failure 38 0.1
Presenting problem not specified in records 5889 20.7 Parkinsons disease 16 0.1

48
 Impact of long-term conditions on psychological therapy outcomes

ordinal regressions standardising on latent variable estimates of Table 3 Estimated coefficients for the seemingly unrelated
depression and anxiety using a Generalised Partial Credit Model regression (SUR) model jointly predicting post-treatment
(uniform and non-uniform DIF), as described by Crane et al.17 depression (PHQ-9) and anxiety (GAD-7) severity
We evaluated differences in Pseudo-R2 values between regression B (s.e.)
models to assess whether the long-term condition categories
Variable PHQ-9 GAD-7
explained a relevant amount of variance in final PHQ-9 and
GAD-7 scores. PHQ-9 (baseline severity) 0.31*** (0.00)
Considering the stepped-care context in which psychological GAD-7 (baseline severity) 0.33*** (0.00)
interventions were provided to this sample, we carried out WSAS (baseline severity) 0.17*** (0.00) 0.13*** (0.00)
secondary analyses to investigate whether patients with long-term Age (years) 70.01** (0.00) 70.01*** (0.00)
conditions may differ in their probability of receiving and Female 70.07 (0.07) 0.02 (0.07)
responding to high-intensity therapy, which would be plausible Unemployed 0.68*** (0.07) 0.54*** (0.07)
given their general higher level of distress and functional Ethnicity (reference category
impairment. To test this hypothesis, we used the same predictors White British)
described above (in the SUR model) in a logistic regression model Dual heritage 0.49* (0.23) 0.56** (0.20)
aiming to predict the likelihood of concluding a treatment episode South Asian 1.29*** (0.18) 1.13*** (0.16)
Black British, African or Caribbean 70.37 (0.25) 70.38 (0.22)
at the higher step of care (high-intensity therapy v. low-intensity
Other 0.53*** (0.15) 0.45*** (0.13)
care as a reference category). In this analysis n = 29 (0.1% of total)
IMD (reference category
case records were excluded because of inputting errors in the
1st IMD quintile)
clinical database, which did not enable us to identify the 2nd quintile 70.77*** (0.11) 70.63*** (0.10)
assignment of individuals to low- or high-intensity steps of care. 3rd quintile 71.11*** (0.11) 70.85*** (0.09)
Furthermore, we repeated the SUR analysis separately analysing 4th quintile 71.36*** (0.11) 71.03*** (0.10)
individuals that finished their treatment episode after accessing 5th quintile 71.75*** (0.12) 71.33*** (0.10)
low- (n = 18 902) or high- (n = 8884) intensity interventions. Medication 70.30 (0.18) 70.20 (0.16)
Since the amount of missing data was only minimal Number of referrals into service 0.59*** (0.03) 0.55*** (0.03)
(nmiss = 683, 2.4% of available cases), no additional imputation Treatment contacts attended 70.25*** (0.01) 70.22*** (0.01)
analyses were undertaken. As expected, most missing data points Mental health (reference
were for the outcome measures (PHQ-9 and GAD-7) and 390 category depressive episode)
individuals did not have valid postcodes documented in clinical Recurrent depression 0.36 (0.19) 0.29 (0.17)
records, so IMD could not be derived. Furthermore, IMD data Mixed anxiety and depression 0.16 (0.10) 0.22* (0.09)
Generalised anxiety disorder 70.56*** (0.14) 70.14 (0.12)
could not be imputed, since it is a geographical rather than an
Social phobia 70.30 (0.27) 70.24 (0.24)
individual characteristic. Panic disorder 70.43* (0.20) 70.12 (0.18)
Agoraphobia 70.09 (0.46) 0.15 (0.41)
Specific phobia 70.23 (0.36) 0.10 (0.31)
OCD 70.25 (0.25) 0.69** (0.22)
Results PTSD 0.55* (0.26) 1.01*** (0.23)
Bereavement 0.76 (0.42) 0.69 (0.38)
SUR equations modelling Eating disorder 1.64*** (0.45) 0.71 (0.40)
As expected, the equations for PHQ-9 and GAD-7 outcomes were Somatoform disorder 71.18* (0.49) 70.66 (0.43)
highly correlated. The correlation between the error terms of the Does not meet CMD criteria 0.38 (0.31) 0.18 (0.27)
two equations was large (r = 0.82) and statistically significant Other diagnosis 0.06 (0.11) 0.16 (0.10)

(BreuschPagan test, w2 = 18 704.17, P50.001). The SUR model LTC (reference category
no self-reported LTC)
explained a moderate amount of variance in post-treatment
Asthma 0.17 (0.14) 0.21 (0.12)
outcomes (PHQ-9 34%; GAD-7 29%). Table 3 presents the Cancer 0.48 (0.55) 0.03 (0.48)
SUR model results in the full sample, which are interpreted Chronic musculoskeletal 1.56*** (0.26) 1.15*** (0.23)
below. In what follows, we refer to combined depression (PHQ-9) COPD 1.92*** (0.51) 1.35** (0.45)
and anxiety (GAD-7) symptoms at the time of the last attended Cardiovascular 0.10 (0.22) 70.05 (0.20)
treatment session as post-treatment distress levels. Diabetes 0.77** (0.29) 0.24 (0.25)
Epilepsy 70.07 (0.43) 0.04 (0.38)
Severe mental health problems 3.36*** (0.51) 2.58*** (0.45)
Other LTC 0.67*** (0.13) 0.52*** (0.11)
Analysis of demographic variables Constant 3.74*** (0.27) 3.15*** (0.24)
Age was significantly correlated with both outcomes, although the Observations 27 815 27 815
association was weak, equivalent to a reduction of 0.01 score R-squared 0.34 0.29
points per year increase. Gender did not correlate with either PHQ-9, Patient Health Questionnaire (depression measure); GAD-7, Generalised
measure. In contrast, unemployment was associated with higher Anxiety Disorder (anxiety measure); WSAS, Work and Social Adjustment Scale;
IMD, index of multiple deprivation (quintile groups); OCD, obsessivecompulsive
average post-treatment distress; with an increase of 0.68 points disorder; PTSD, post-traumatic stress disorder; CMD, common mental disorder;
for PHQ-9 and 0.54 points for GAD-7. Patients from South Asian LTC, long-term condition; COPD, chronic obstructive pulmonary disease.
*P50.05; **P50.01; ***P50.001.
backgrounds, dual heritage and other ethnicities also had higher
average post-treatment distress in comparison with patients from
White British backgrounds. The IMD quintiles also show Analysis of treatment-related variables
significant associations in the expected direction: patients living Differences attributable to taking medication were small and not
in more socioeconomically deprived areas (as defined by their statistically significant. The number of referrals for psychological
home postcode) tended to have greater symptom severity at the treatment (over the past 5 years) was significantly associated with
end of treatment, and there was an increasing trend in mean post-treatment outcomes. Each additional referral for care
post-treatment scores for each quintile of deprivation. predicted an increase of 0.59 points for PHQ-9 and 0.55 for

49
Delgadillo et al

GAD-7. Since up to 11 referrals were observed in the data-set, this RMSEGAD-7 = 5.14). The effect sizes for those five long-term
could account for substantially greater post-treatment symptom condition categories, associated with poorer outcomes, range from
severity for patients with multiple treatment episodes. A greater small (around d = 0.20) to medium (around 0.50; according to
number of treatment contacts (in the index treatment episode) Cohen18).
was associated with lower average post-treatment distress; decreasing
this by 0.25 PHQ-9 points per contact attended (0.22 for GAD-7).
Some differences between diagnostic categories were observed. Sensitivity and secondary analyses
A patient with a diagnosis of depression (the reference category) will Our sensitivity analyses found no evidence of DIF on PHQ-9
finish therapy with a predicted PHQ-9 score of 9.69 (s.e. = 0.08, or GAD-7 across long-term condition groups. The maximal
marginal mean) and GAD-7 score of 8.45 (s.e. = 0.07). Some difference in Pseudo-R2 values across regression models was
diagnoses were associated with higher post-treatment distress 0.001, which is far smaller than the recommended cut-off of
in comparison with the above reference scores; these were 0.035,17 indicating that only very small amounts of variance in
obsessivecompulsive disorder (GAD-7), post-traumatic stress item responses were related to specific long-term conditions
disorder (PHQ-9 and GAD-7) and eating disorders (PHQ-9). (online supplement DS1 and Tables DS1DS8).
Diagnoses associated with lower predicted distress levels The SUR model including a dichotomous variable (long-term
compared with depression were generalised anxiety disorder condition v. no long-term condition) resulted in an adjusted mean
(PHQ-9), panic disorder (PHQ-9) and somatoform disorders difference of 0.57 (s.e. = 0.09, P50.001) for the PHQ-9 and 0.42
(PHQ-9). As expected, baseline severity (PHQ-9, GAD-7) and for GAD-7 (s.e. = 0.08, P50.001), which speaks for a general
functional impairment (WSAS) measures were significantly disadvantage of this population with long-term conditions after
correlated with post-treatment outcomes, such that individuals controlling for potentially confounding variables. However, this
who were more severely impaired were expected to have higher effect is rather small (ESPHQ = 0.10 and ESGAD = 0.08; coefficients
post-treatment distress severity. standardised on regression RMSE) and therefore modelling post-
treatment outcomes for each specific long-term condition is more
informative. When we repeated the SUR analysis in the separate
Analysis of long-term conditions data samples of individuals that finished treatment after low- or
After controlling for all of the above demographic and clinically high-intensity interventions, the results (online supplement DS2
relevant variables, five long-term condition categories were and Table DS9) were largely consistent with the main SUR model
associated with higher post-treatment distress levels: chronic shown in Table 3. The only differences were found for people with
musculoskeletal problems, COPD, severe mental health problems, COPD and diabetes. COPD continued to be associated with
diabetes, and other non-specified conditions. Figure 1 presents higher post-treatment PHQ-9 (B = 3.81, P50.001) and GAD-7
predicted marginal post-treatment outcome scores and Sidak- (B = 2.34, P50.01) scores after high-intensity interventions, but
adjusted 95% confidence intervals; this shows that estimated not after low-intensity ones (PHQ-9: B = 0.90, P40.05; GAD-7:
post-treatment means differed substantially across the categories B = 0.75, P40.05). Diabetes continued to be associated with higher
of long-term conditions. The figure also presents the estimated post-treatment PHQ-9 (B = 1.30, P50.01) but not with GAD-7
treatment effect sizes corresponding to each long-term condition (B = 0.50, P40.05) scores after high-intensity interventions; no
category, by comparison with the reference group without such associations were found for individuals that only accessed
any self-reported long-term conditions (and in relation to non- low-intensity interventions (PHQ-9: B = 0.43, P40.05; GAD-7:
explained variance in the regression model; RMSEPHQ-9 = 5.82; B = 0.00, P40.05).

(a) (b)

21 27
Predicted post-treatment GAD-7

Predicted post-treatment PHQ-9

18 24
d= d= d= d= d= d= d= d= d= d= 21 d= d= d= d= d= d= d= d= d= d=
15 0.04 0.01 0.22a 0.26a 70.01 0.05 0.01 0.50a 0.10a 0.03 0.08 0.27a 0.33a 0.02 0.13a 70.01 0.58a 0.11a
18
12 15

9 12

9
6
6
3
3
0 0
es

es
sk ic

sk ic
D
a

a
s

s
ar

y
C

C
cu Ch r

er

al
al

r
ar
D
e

er

er
m

m
ps

ps
he

he
OP
lo ron

lo ron
LT

LT
et

et
ul

et

et
nc

nc

OP

ul
th

th
rd

rd
le

le
Ot

Ot
sc
el

el
ab

ab
C

sc
No

No
Ca

Ca

cu Ch
As

As
so

so
C
pi

i
Ep
va

va
Di

Di
E

di

di
io

io
tic

tic
rd

rd
us

us
Ca

ho

ho
Ca
m

m
yc

yc
Ps

Ps

Fig. 1 Predicted post-treatment anxiety (Generalised Anxiety Disorder scale (GAD-7)) and depression (Patient Health Questionnaire
(PHQ-9)) scores across long-term condition (LTC) groups
Marginal post-treatment scores derived from seemingly unrelated regression model (Table 3) jointly predicting GAD-7 (a) and PHQ-9 (b) for a patient starting with mean age = 38.38,
PHQ-9 = 15.17 and GAD-7 = 13.38 scores. The y-axes of both plots present the potential range in scores and the effect sizes (Cohens d) are based on comparisons with the no
long-term condition category and the root mean square error of the respective part of the regression model.
a. Coefficient significant (see Table 3). COPD, chronic obstructive pulmonary disease; Other, other non-specified LTC.

50

Table 4 Probability of accessing high-intensity therapy for
patients with long-term conditions a
Long-term condition Odds ratio (95% CI)
Asthma 1.23** (1.091.39)
Cancer 1.66* (1.032.65)
Chronic musculoskeletal 1.47** (1.171.84)
Chronic obstructive pulmonary disease 1.60* (1.042.47)
Cardiovascular 1.15 (0.941.40)
Diabetes 1.39* (1.071.79)
Epilepsy 0.99 (0.681.44)
Severe (psychotic) mental health problems 1.00 (0.641.57)
Other non-specified long-term condition 1.30*** (1.171.46)
a. All coefficients controlled for demographic and treatment-related variables as for
the main analysis (see Table 3).
*P50.05; **P50.01; ***P50.001.
Table 4 shows the results of the logistic regression model
predicting the likelihood of receiving high-intensity therapy. As
expected, patients with most long-term conditions had a
significantly higher probability of receiving high-intensity
interventions (odds ratios (OR) = 1.231.66; all P50.05). The
only exceptions to this were found for patients with cardiovascular
conditions, epilepsy and severe (psychotic) mental disorders, who
were no more likely to receive high-intensity therapy in
comparison with those without any self-reported long-term
conditions.
Discussion
The overall goal of this study was to compare the effects of
routinely delivered psychological care for patients with and
without self-reported long-term conditions. We found that
patients with certain long-term conditions tend to finish
psychological treatment with greater depression and anxiety
severity. In comparison with patients without long-term
conditions, this trend was statistically significant for musculoskeletal
problems (effect sizes d = 0.220.27), COPD (d = 0.260.33),
diabetes (d = 0.050.13), severe mental health problems
(d = 0.500.58) and other non-specified conditions (d = 0.10
0.11). These findings converge with some observations of small to
moderate effect sizes reported by meta-analyses of psychological
interventions.6,1921 Furthermore, patients with most types of
self-reported long-term conditions (except cardiovascular
conditions, epilepsy and severe mental disorders) were
significantly more likely to receive more intensive and costly
psychological interventions, consistent with their higher level of
impairment and symptom severity. Accessing high-intensity therapy
continued to be associated with higher average post-treatment
distress in secondary analyses, indicating that patients with
long-term conditions are not necessarily better off after high-
intensity (v. low-intensity) care in this primary care setting. We
noted that, in particular, severe mental health problems (such as
psychotic disorders) tended to be strongly associated with poorer
outcomes, although the prevalence of such conditions was very
small (0.5%) given the services remit to offer treatment for
common mental health problems. This evidence may indicate that
the intensity and type of interventions offered in this primary care
setting are clearly inadequate to improve psychological distress
symptoms in people with a history of severe and enduring (i.e.
psychotic) mental disorders. It is also plausible that those with
severe mental disorders were identified as such during the early
phases of low-intensity interventions and appropriately referred
onwards to secondary care/psychiatric services; hence explaining
Impact of long-term conditions on psychological therapy outcomes
why these individuals were not more likely to access high-intensity
therapies in this primary care setting.
A range of other demographic and clinical factors were also
associated with post-treatment outcomes. Baseline severity and
impairment measures, employment status, socioeconomic
deprivation and age have been shown to predict outcomes in
comparable clinical samples and settings.2224 In addition, the
present results also indicate that patients from certain ethnic
backgrounds (South Asian, dual heritage) and diagnostic groups
(obsessivecompulsive disorder, post-traumatic stress disorder
and eating disorders) may be at increased risk of poor treatment
outcomes.
The overall prevalence of self-reported long-term conditions
in this cohort of patients receiving psychological treatment
(23.2%) was comparable with general adult population estimates
(20%) from England.1 Asthma was the most prevalent (identifiable)
long-term condition (around 6% of patients); which may reflect
the common coexistence of this condition with anxiety-related
problems. However, a comparison of specific categories reveals a
disproportionately small representation of certain conditions such
as hypertension (1.7% in our sample v. 14.3% in the general
population), musculoskeletal problems (1.8% v. 14.0%), diabetes
(1.5% v. 3.8%), heart failure (0.1% v. 2.0%) and COPD (0.5% v.
1.8%). These discrepancies possibly indicate that patients with
certain conditions are much less likely to access psychological care;
for example, the large discrepancy for hypertension indicates a
ratio of 1:12. This discrepancy might in some cases be explained
by the existence of specialist teams for people with long-term
conditions; for example, the local area for this cohort of patients
had two specialist musculoskeletal treatment services. However,
it is possible that some patients with long-term conditions
commonly treated in primary care clinics may be inadequately
screened or seldom referred for mental health treatment. Recent
research indicates that this may be the case for some patients with
coronary heart disease and diabetes.25
Strengths and limitations
This large (n = 28 498) naturalistic cohort was adequately powered
to assess the predictive value of multiple demographic and clinical
variables, which was optimally modelled using joint-prediction of
correlated outcomes (PHQ-9, GAD-7). Our analyses additionally
contained another layer of robustness tests. It is important to
establish measurement invariance to have confidence that
outcome questionnaires can be interpreted similarly across
different patient groups.26,27 Our sensitivity analyses found no
evidence of DIF in PHQ-9 or GAD-7, which indicates that results
of these outcome measures have the same meaning across long-
term conditions and non-long-term condition groups in this
sample. This finding supports the notion that post-treatment
outcome differences between long-term condition groups are
likely to represent actual differences in psychological distress,
rather than measurement error or confounding of long-term
condition symptoms with mental health symptoms.
Some limitations to note include the reliance on self-report
of long-term conditions, as is common in large cohort and
epidemiological studies.1,28 In particular, we did not have more
specific information about the types of long-term condition of
patients who endorsed the other category in assessment records.
Presenting problem categories for mental health issues were also
likely to be error prone, since these were ascertained using brief
screening measures12 rather than structured diagnostic interviews.
Future studies that gather long-term condition diagnoses recorded
in medical records and structured diagnostic interviews may
render more precise outcome prediction estimates. A further
51
Delgadillo et al
caveat is that the data from this study, albeit large, came from a
single site in the north of England. Future replication studies using
data from similar stepped-care services in other regions would
enable us to assess the extent to which these findings are generalisable.
Implications for clinical practice
The impact of, and need for, integrated mental health services
has often been discussed from a medical perspective, i.e. how
psychological professionals could be brought into medical
contexts.29,30 Specialist medical knowledge about long-term
conditions is a key element of success and the integration of
medical expertise within mental health services could also help to
improve treatment outcomes in these settings. Our observations
raise questions about the effectiveness of routinely delivered
stepped-care psychological treatments for people with comorbid
diabetes, COPD and chronic pain. These results also mirror findings
from research into health-related quality of life (QOL), where it is
also found that certain long-term conditions (such as chronic pain)
and especially having multiple long-term conditions can considerably
undermine QOL and exacerbate psychological distress.31 Such studies
highlight the importance of multidisciplinary care aiming to target
multiple facets of well-being, adjustment and QOL. It may be
particularly important to offer integrated multidisciplinary care
for people with specific conditions described above. For example,
collaborative care interventions can enhance self-management of
depression symptoms for patients with diabetes and coronary
heart disease.29 Overall, we conclude that standard stepped-care
interventions are insufficient to support patients with multi-
morbidity, especially if delivered in isolation from other healthcare
specialists. Our observations concur with recent calls for closer
integration of physical and mental healthcare.32
Healthcare economies and policy-makers should systematically
investigate the prevalence of long-term conditions in people using
mental health services. This information could be crucial to design
and deliver more integrated care for patients with long-term
conditions, but can also serve to understand how the demographic
and clinical profile of local populations could have an impact on
service outcomes. This aspect is still underexplored, especially
when thinking about new benchmarking models and quality
indicators within primary care psychological services.33
Jaime Delgadillo, PhD, Leeds Community Healthcare NHS Trust and Department
of Health Sciences, University of York, York, UK; Alexander Dawson, Leeds
Community Healthcare NHS Trust, Leeds; Simon Gilbody, DPhil, FRCPsych, Jan R.
Bohnke, Dipl Psych, Dr rer nat, Hull York Medical School, and Department of Health
Sciences, University of York, York, UK.
Correspondence: Jaime Delgadillo, Clinical Psychology Unit, Department of
Psychology, University of Sheffield, Western Bank, Sheffield S10 2TN, UK. Email:
jaime.delgadillo@nhs.net
First received 8 Jun 2016, final revision 25 Jul 2016, accepted 30 Jul 2016
Funding
This study was funded by Leeds Community Healthcare NHS Trust and commissioned by
the Long Term Conditions Project Team, Leeds, UK, January 2016.
Acknowledgements
We thank Peter A. Coventry for reviewing an earlier version of the study manuscript and
for helpful advice.
References
1 Weich S, Bebbington P, Rai D, Stranges S, McBride O, Spiers N, et al. The
population impact of common mental disorders and long-term physical
conditions on disability and hospital admission. Psychol Med 2013; 43:
92131.
52
2 Department of Health. Long Term Conditions Compendium of Information
(3rd edn). Department of Health, 2012.
3 Barnett K, Mercer SW, Norbury M, Watt G, Wyke S, Guthrie B. Research
paper. Epidemiology of multimorbidity and implications for health care,
research, and medical education: a cross-sectional study. Lancet 2012; 380:
3743.
4 Department of Health. Talking Therapies: A Four-Year Plan of Action.
Department of Health, 2011.
5 Dickens C, Cherrington A, Adeyemi I, Roughley K, Bower P, Garrett C,
et al. Characteristics of psychological interventions that improve
depression in people with coronary heart disease: a systematic review
and meta-regression. Psychosom Med 2013; 75: 21121.
6 Williams ACD, Eccleston CM, Morley S. Psychological therapies for the
management of chronic pain (excluding headache) in adults. Cochrane
Database System Rev 2012; 11: CD007407.
7 Clark DM, Layard R, Smithies R, Richards DA, Suckling R, Wright B. Improving
access to psychological therapy: initial evaluation of two UK demonstration
sites. Behav Res Ther 2009; 47: 91020.
8 National Institute for Health and Care Excellence. Common Mental Health
Disorders: Identification and Pathways to Care. CG123. National Institute for
Health and Care Excellence, 2011 (http://www.nice.org.uk/guidance/CG123).
9 Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief depression
severity measure. J Gen Intern Med 2001; 16: 60613.
10 Kroenke K, Spitzer RL, Williams JBW, Monahan PO, Lowe B. Anxiety disorders
in primary care: prevalence, impairment, comorbidity, and detection. Ann
Intern Med 2007; 146: 31725.
11 Mundt JC, Marks IM, Shear MK, Greist JM. The Work and Social Adjustment
Scale: a simple measure of impairment in functioning. Br J Psychiatry 2002;
180: 4614.
12 IAPT National Programme Team. The IAPT Data Handbook: Guidance on
Recording and Monitoring Outcomes to Support Local Evidence-Based
Practice. Version 2.0.1. National IAPT Programme Team, 2011 (http://
www.iapt.nhs.uk/silo/files/iapt-data-handbook-v2.pdf)
13 Department for Communities and Local Government. The English Indices of
Deprivation 2010. Department for Communities and Local Government, 2011.
14 Bohnke JR, Lutz W, Delgadillo J. Negative affectivity as a transdiagnostic
factor in patients with common mental disorders. J Affect Disord 2014; 166:
2708.
15 Zellner A. An efficient method of estimating seemingly unrelated regressions
and tests for aggregation bias. J Am Stat Assoc 1962; 57: 34868.
16 Keshavarzi S, Ayatollahi SMT, Zare N, Farkhondeh S. Quality of life of
childbearing age women and its associated factors: an application of
seemingly unrelated regression (SUR) models. Qual Life Res 2013; 22:
125563.
17 Crane PK, Gibbons LE, Ocepek-Welikson K, Cook K, Cella D, Narasimhalu K,
et al. A comparison of three sets of criteria for determining the presence of
differential item functioning using ordinal logistic regression. Qual Life Res
2007; 16: 6984.
18 Cohen J. Statistical Power Analysis for the Behavioural Sciences (2nd edn).
Erlbaum, 1988.
19 Coventry PA, Gellatly JL. Improving outcomes for COPD patients with
mild-to-moderate anxiety and depression: a systematic review of cognitive
behavioural therapy. Br J Health Psychol 2008; 13: 381400.
20 Panagioti M, Scott C, Blakemore A, Coventry PA. Overview of the prevalence,
impact, and management of depression and anxiety in chronic obstructive
pulmonary disease. Int J Chron Obstruct Pulmon Dis 2014; 9: 1289306.
21 van der Feltz-Cornelis CM, Nuyen J, Stoop C, Chan J, Jacobson AM, Katon W,
et al. Effect of interventions for major depressive disorder and significant
depressive symptoms in patients with diabetes mellitus: a systematic review
and meta-analysis. Gen Hosp Psychiatry 2010; 32: 38095.
22 Delgadillo J, Asaria M, Ali S, Gilbody S. On poverty, politics and psychology:
the socioeconomic gradient of mental healthcare utilisation and outcomes.
Br J Psychiatry 2016; 209: 42930.
23 Delgadillo J, Moreea O, Lutz W. Different people respond differently to
therapy: a demonstration using patient profiling and risk stratification. Behav
Res Ther 2016; 79: 1522.
24 Firth N, Barkham M, Kellett S, Saxon D. Therapist effects and moderators
of effectiveness and efficiency in psychological wellbeing practitioners:
a multilevel modelling analysis. Behav Res Ther 2015; 69: 5462.
25 Coventry PA, Hays R, Dickens C, Bundy C, Garrett C, Cherrington A, et al.
Talking about depression: a qualitative study of barriers to managing
depression in people with long term conditions in primary care. BMC Fam
Pract 2011; 12: 10.
 Impact of long-term conditions on psychological therapy outcomes

26 Bohnke JR, Croudace TJ. Factors of psychological distress: clinical value,
measurement substance, and methodological artefacts. Soc Psychiatry
Psychiatr Epidemiol 2015; 50: 51524.
27 Teresi JA, Ocepek-Welikson K, Kleinman M, Eimicke JP, Crane PK, Jones RN,
et al. Analysis of differential item functioning in the depression item bank
from the patient reported outcome measurement information system
(PROMIS): an item response theory approach. Psychol Sci Q 2009; 51:
14880.
28 Patten SB, Beck CA, Kassam A, Williams JV, Barbui C, Metz LM. Long-term
medical conditions and major depression: strength of association for specific
conditions in the general population. Can J Psychiatry 2005; 50: 195202.
29 Coventry P, Lovell K, Dickens C, Bower P, Chew-Graham C, McElvenny D,
et al. Integrated primary care for patients with mental and physical
multimorbidity: cluster randomised controlled trial of collaborative care for
patients with depression comorbid with diabetes or cardiovascular disease.
BMJ 2015; 350: h638.
30 McDaniel SH, deGruy III FV. An introduction to primary care and psychology.
Am Psychol 2014; 69: 32531.
31 Tyack Z, Frakes KA, Barnett A, Cornwell P, Kuys S, McPhail S. Predictors
of health-related quality of life in people with a complex chronic disease
including multimorbidity: a longitudinal cohort study. Qual Life Res 2016;
25: 257992.
32 Naylor C, Das P, Ross S, Honeyman M, Thompson J, Gilburt H. Bringing
Together Physical and Mental Health: A New Frontier for Integrated Care.
Kings Fund, 2016.
33 Delgadillo J, McMillan D, Leach C, Lucock M, Gilbody S, Wood N.
Benchmarking routine psychological services: a discussion of challenges and
methods. Behav Cogn Psychother 2014; 42: 1630.

Research of Hope
poems
by Sally Fox
doctors
My perception of safety has always been skewed
My defensive shyness often construed as rudeness
The freezing sensation I feel inside
when introduced to a new person
Im searching their features for a sense of familiar,
looking for something kind or similar
But danger is always omnipresent.
It makes me feel awkward and hesitant,
resistant to friendship.

Then at a PD conference Jeremy Hall spoke

of the research hed done
with Merrick Pope and another.
MRI scans of borderline brains
their findings seem to proclaim
Im not a social moron after all;
its down to that concentration of neurons
that almond cluster they call the amygdala
in the pre-frontal cortex.

Thats the reason for my social awkwardness.

Its hypersensitivity makes so much sense to me
born of childhood adversity
confused facial emotion recognition
affecting my social cognition.
This science of magnetic resonance imaging
is so bloody validating!
At long last some Dignity and Hope.
Thank you: Nicol, Hall and Pope!

Selected by Femi Oyebode. From Stigma & Stones: Living with a Diagnosis of BPD, poems by Sally Fox & Jo McFarlane.
B Sally Fox. Reprinted with permission.

Through their collection Stigma & Stones, writers/performers/partners Sally Fox and Jo McFarlane seek to promote
understanding, improve treatment and reduce the stigma of living with a diagnosis of BPD.
The British Journal of Psychiatry (2017)
210, 53. doi: 10.1192/bjp.bp.116.181719

53
Data supplement to Delgadillo et al. Impact of long-term medical conditions on the
outcomes of psychological therapy for depression and anxiety. Br J Psychiatry doi:
10.1192/bjp.bp.116.189027

Online supplement DS1

Differential item functioning (DIF) analysis

The following eight tables present the results of the analysis for differential item functioning (DIF) in
more detail. Details on the steps of the procedure will be presented using the example of the first item
of the PHQ-9 (PHQ1) at the first available assessment (tables DS1 and DS2). All models were
estimated in Stata 14 (StataCorp, College Station, Texas, 2015).

For unidimensional and unbiased instruments it would be expected that the response of a single item
can be predicted to a large extent with an estimate of the latent variable that the instrument is
supposed to measure. DIF would be deemed to present in a specific item, if after controlling for the
latent variable other variables would still be predictive of the item response. Approaches based on
logistic regression models17 operationalise this straight forwardly: In a first step, the item to be tested
is regressed upon an estimate of the latent trait and in a second step the variable is added that is tested
for potential bias. In our example, the estimate of the latent trait when predicting PHQ1 was generated
with a Generalised Partial Credit Model34 for items PHQ2-PHQ9. In the first step (columns "Trait" in
all tables), only this estimate was used to predict the response to item PHQ9 with an ordinal logistic
regression model. The tables present the Pseudo-R (e.g., .242 for PHQ1, table DS1) as well as the
result of the comparison between this model and the baseline model without any predictors (Wald-
df=1 = 12630.59; p<.001; table DS1). This comparison is highly significant and a relevant amount of
variance in PHQ1 responses is explained, which indicates that items and trait are highly correlated as
to be expected. This result is found for all items (across all tables).

In the second step of this approach, the grouping variable is added (columns "Trait + LTC" in all
tables), in our case "no LTC vs. any LTC" (table DS1). The comparison test compares now the model
including two predictors with the model with only the estimate of the latent variable (Wald-
df=1 = 11.82; p<.001; table DS1). And while this test is highly significant (suggesting the potential of
DIF), the main interpretation for the relevance of DIF is the comparison of the Pseudo-R values of the
two models. Generally a cut-off of >0.035 is judged to be indicative of DIF that could potentially
affect group comparisons17. In this case, as well as for all other comparisons presented between these
two models the differences in Pseudo-R are well below this cut off, suggesting that the significant
effects found in some comparisons are due to the statistical power of the large sample size.

The two steps so far have tested whether there is a constant difference in the probability of endorsing
certain symptoms, in this case whether for example the group with any LTC shows, after controlling
for distress, a higher or lower probability across the whole severity spectrum. This is usually called
"uniform DIF".35 But one could also imagine the case that the severity and LTC presence interact, e.g.,
that patients with higher levels of distress and an LTC are more likely to respond in the top categories
of PHQ1. Such an effect would be called "non-uniform" DIF35 and is tested for by adding in a third
model the interaction term between the latent variable estimate and the grouping variable (columns
"Trait X LTC" in all tables). The model test compares now the model including the interaction term
with the model that contains only the latent variable estimate and the grouping variable. In the chosen
example, this test is not significant (Wald-df=1 = 3.26; p=.07; table DS1), which indicates that the
interaction term does not add any information above and beyond the independent effects of the two
variables. Again, the relevant result would be if a difference in Pseudo-R values would be detected
between this model and either the "Trait" or the "Trait + LTC" models, which is not the case for any
of the tested items.

Per outcome measure (PHQ-9 and GAD-7) and assessment (first and last available) two tables are
presented each: the first of these shows the results for no "LTC vs any LTC", while the second one
presents the analyses with the LTC categories as used in the regression analysis presented in the main
body of the research manuscript.

Overall, the increases in Pseudo-R values for all items and all comparisons are very low, indicating
that DIF is unlikely to have an impact on the results at either first or last assessments with both
instruments

Table DS1 Results for DIF analyses of the PHQ-9 responses at the first recorded assessment
comparing any LTC vs no LTC

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCa Trait X LTCa
12630.59 11.82 3.26
PHQ1 .242 .243 .243
(p<.001) (p<.001) (p=0.07)
12699.73 1.01 0.33
PHQ2 .275 .275 .275
(p<.001) (p=0.32) (p=0.57)
7633.45 23.23 0.07
PHQ3 .136 .136 .136
(p<.001) (p<.001) (p=0.79)
8933.49 54.45 2.31
PHQ4 .170 .171 .171
(p<.001) (p<.001) (p=0.13)
8318.27 5.87 0.45
PHQ5 .133 .133 .133
(p<.001) (p=0.02) (p=0.50)
9906.78 22.28 0.09
PHQ6 .180 .180 .180
(p<.001) (p<.001) (p=0.76)
9677.69 0.14 0.31
PHQ7 .158 .158 .158
(p<.001) (p=0.71) (p=0.58)
6651.89 13.84 4.11
PHQ8 .106 .106 .106
(p<.001) (p<.001) (p=0.04)
6005.02 9.36 0.25
PHQ9 .121 .121 .121
(p<.001) (p=0.002) (p=0.62)
Note. aDegrees of freedom = 1
Table DS2: Results for DIF analyses of the PHQ-9 responses at the first recorded assessment
comparing all LTC groups (no LTC as reference category)

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCb Trait X LTCb
12630.59 48.10 21.85
PHQ1 .242 .243 .243
(p<.001) (p<.001) (p=0.01)
12699.73 7.48 15.45
PHQ2 .275 .275 .275
(p<.001) (p=0.59) (p=0.08)
7633.45 31.40 5.26
PHQ3 .136 .137 .137
(p<.001) (p<.001) (p=0.81)
8933.49 71.60 13.35
PHQ4 .170 .171 .171
(p<.001) (p<.001) (p=0.15)
8318.27 17.21 9.11
PHQ5 .133 .133 .133
(p<.001) (p=0.05) (p=0.43)
9906.78 61.89 8.90
PHQ6 .180 .181 .181
(p<.001) (p<.001) (p=0.45)
9677.69 15.97 17.58
PHQ7 .158 .158 .158
(p<.001) (p=0.07) (p=0.04)
6651.89 28.87 9.83
PHQ8 .106 .106 .107
(p<.001) (p<.001) (p=0.36)
6005.02 25.14 8.98
PHQ9 .121 .121 .121
(p<.001) (p=0.003) (p=0.44)
Note. aDegrees of freedom = 1; bDegrees of freedom = 9

Table DS3: Results for DIF analyses of the PHQ-9 responses at the last recorded assessment
comparing any LTC vs no LTC

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCa Trait X LTCa
14115.43 1.24 8.74
PHQ1 .407 .407 .407
(p<.001) (p=0.27) (p=0.003)
14438.18 0.06 5.36
PHQ2 .396 .396 .397
(p<.001) (p=0.81) (p=0.02)
12792.29 26.68 5.21
PHQ3 .252 .252 .253
(p<.001) (p<.001) (p=0.02)
13947.03 38.67 0.36
PHQ4 .298 .298 .298
(p<.001) (p<.001) (p=0.55)
11793.81 6.03 3.97
PHQ5 .247 .248 .248
(p<.001) (p=0.01) (p=0.05)
13498.00 16.91 0.01
PHQ6 .317 .317 .317
(p<.001) (p<.001) (p=0.94)
13049.03 0.21 2.88
PHQ7 .306 .306 .306
(p<.001) (p=0.65) (p=0.09)
9353.13 9.42 0.00
PHQ8 .244 .244 .244
(p<.001) (p=0.002) (p=0.97)
6238.94 2.70 0.02
PHQ9 .253 .253 .253
(p<.001) (p=0.10) (p=0.89)
Note. aDegrees of freedom = 1
Table DS4: Results for DIF analyses of the PHQ-9 responses at the last recorded assessment
comparing all LTC groups (no LTC as reference category)

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCb Trait X LTCb
PHQ1 .407 .407 .407 14115.43 27.04 10.36
(p<.001) (p=0.001) (p=0.32)
PHQ2 .396 .397 .397 14438.18 12.14 8.59
(p<.001) (p=0.21) (p=0.48)
PHQ3 .252 .253 .253 12792.29 36.98 12.32
(p<.001) (p<.001) (p=0.20)
PHQ4 .298 .298 .298 13947.03 49.56 9.67
(p<.001) (p<.001) (p=0.38)
PHQ5 .247 .248 .248 11793.81 16.95 18.08
(p<.001) (p=0.05) (p=0.03)
PHQ6 .317 .318 .318 13498.00 39.79 11.34
(p<.001) (p<.001) (p=0.25)
PHQ7 .306 .306 .306 13049.03 17.23 7.31
(p<.001) (p=0.04) (p=0.61)
PHQ8 .244 .245 .245 9353.13 31.10 22.32
(p<.001) (p<.001) (p=0.01)
PHQ9 .253 .253 .254 6238.94 8.90 10.83
(p<.001) (p=0.45) (p=0.29)
Note. aDegrees of freedom = 1; bDegrees of freedom = 9

Table DS5: Results for DIF analyses of the GAD-7 responses at the first recorded assessment
comparing any LTC vs no LTC

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCa Trait X LTCa
11000.50 0.14 3.19
GAD1 .238 .238 .238
(p<.001) (p=0.71) (p=0.07)
12639.41 0.40 12.68
GAD2 .332 .332 .333
(p<.001) (p=0.53) (p=0.0004)
12306.44 1.22 0.09
GAD3 .329 .329 .329
(p<.001) (p=0.27) (p=0.76)
11352.83 15.96 0.60
GAD4 .218 .218 .218
(p<.001) (p<.001) (p=0.44)
7712.67 20.48 1.32
GAD5 .118 .118 .118
(p<.001) (p<.001) (p=0.25)
5840.43 7.26 4.74
GAD6 .088 .088 .088
(p<.001) (p=0.007) (p=0.03)
9292.46 7.37 1.93
GAD7 .148 .148 .148
(p<.001) (p=0.007) (p=0.16)
Note. aDegrees of freedom = 1
Table DS6: Results for DIF analyses of the GAD-7 responses at the first recorded assessment
comparing all LTC groups (no LTC as reference category)

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCb Trait X LTCb
11000.50 9.25 9.77
GAD1 .238 .238 .238
(p<.001) (p=0.41) (p=0.37)
12639.41 6.10 24.19
GAD2 .332 .333 .333
(p<.001) (p=0.73) (p=0.004)
12306.44 16.75 3.78
GAD3 .329 .330 .330
(p<.001) (p=0.05) (p=0.93)
11352.83 31.71 4.64
GAD4 .218 .219 .219
(p<.001) (p<.001) (p=0.86)
7712.67 44.53 9.15
GAD5 .118 .119 .119
(p<.001) (p<.001) (p=0.42)
5840.43 35.15 10.70
GAD6 .088 .089 .089
(p<.001) (p<.001) (p=0.30)
9292.46 24.97 6.27
GAD7 .148 .148 .148
(p<.001) (p=0.003) (p=0.71)
Note. aDegrees of freedom = 1; bDegrees of freedom = 9

Table DS7: Results for DIF analyses of the GAD-7 responses at the last recorded assessment
comparing any LTC vs. no LTC

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCa Trait X LTCa
14742.80 0.79 2.89
GAD1 .384 .384 .384
(p<.001) (p=0.37) (p=0.09)
14700.92 2.35 1.36
GAD2 .461 .461 .461
(p<.001) (p=0.13) (p=0.24)
14787.63 0.66 0.92
GAD3 .455 .455 .455
(p<.001) (p=0.42) (p=0.34)
14419.29 3.42 2.38
GAD4 .343 .343 .344
(p<.001) (p=0.06) (p=0.12)
11245.83 15.78 6.72
GAD5 .260 .260 .260
(p<.001) (p<.001) (p=0.01)
12058.38 1.36 2.18
GAD6 .234 .235 .235
(p<.001) (p=0.24) (p=0.14)
12396.65 1.48 0.00
GAD7 .288 .288 .288
(p<.001) (p=0.22) (p=0.95)
Note. aDegrees of freedom = 1
Table DS8: Results for DIF analyses of the GAD-7 responses at the last recorded assessment
comparing all LTC groups (no LTC as reference category)

Pseudo-R Wald- (p-value)

Trait Trait + LTC Trait X LTC Traita Trait + LTCb Trait X LTCb
14742.80 15.70 11.12
GAD1 .384 .384 .384
(p < .001) (p=0.07) (p=0.27)
14700.92 12.24 18.43
GAD2 .461 .461 .461
(p < .001) (p=0.20) (p=0.03)
14787.63 4.39 4.14
GAD3 .455 .455 .455
(p < .001) (p=0.88) (p=0.90)
14419.29 34.03 9.38
GAD4 .343 .344 .344
(p < .001) (p<0.001) (p=0.40)
11245.83 39.82 15.66
GAD5 .260 .261 .261
(p < .001) (p<.001) (p=0.07)
12058.38 32.08 10.45
GAD6 .234 .235 .235
(p < .001) (p<.001) (p=0.32)
12396.65 9.33 13.37
GAD7 .288 .288 .288
(p < .001) (p=0.41) (p=0.15)
Note. aDegrees of freedom = 1; bDegrees of freedom = 9

Online supplement DS2

Assessing the impact of LTC on post-treatment depression (PHQ-9) and anxiety (GAD-7): Low
versus high intensity treatment pathways

To test for potential interaction effects of the intensity of treatment provided and the LTCs, we re-ran
the model described in table 3 in the main body of this manuscript separately for cases that finished
their treatment pathway after low intensity interventions (n = 18 902) and high intensity interventions
(n = 8884). The following describes only differences that were found in significance or direction
compared to the main analysis (detailed results in table DS9). Focusing only on the coefficients for the
LTCs, we found very few changes compared to the analysis on the full sample. Both, COPD and
Diabetes were only correlated with higher PHQ-9 (and for COPD: GAD-7) scores in patients finishing
their treatment pathway after high intensity interventions. All other relationships remained the same.
Table DS9: Estimated LTC coefficients for the SUR model jointly predicting post-treatment
depression (PHQ-9) and anxiety (GAD-7) severity for low and high intensity treatment
pathways

Low intensity pathway High intensity pathway

LTC PHQ-9 GAD-7 PHQ-9 GAD-7
(SE) (SE) (SE) (SE)

Asthma 0.18 0.26 0.16 0.10

(0.17) (0.15) (0.24) (0.22)
Cancer -0.00 -0.46 0.88 0.45
(0.64) (0.57) (0.99) (0.88)
Chronic Musculoskeletal 1.84*** 1.24*** 1.01* 0.86*
(0.32) (0.28) (0.46) (0.40)
COPD 0.90 0.75 3.81*** 2.34**
(0.60) (0.53) (0.93) (0.83)
Cardiovascular 0.05 -0.09 0.23 -0.00
(0.26) (0.23) (0.43) (0.38)
Diabetes 0.43 0.00 1.30** 0.50
(0.34) (0.30) (0.51) (0.45)
Epilepsy -0.19 -0.05 0.25 0.33
(0.50) (0.44) (0.80) (0.71)
Severe Mental (psychotic) 2.93*** 2.08*** 4.13*** 3.55***
disorder (0.61) (0.54) (0.89) (0.79)
Other LTC 0.65*** 0.57*** 0.67** 0.33
(0.15) (0.13) (0.23) (0.20)

Constant 4.31*** 3.70*** 4.82*** 4.22***

(0.32) (0.28) (0.49) (0.44)
Observations 18 902 18 902 8884 8884
R-squared 0.38 0.33 0.31 0.27
Notes: SUR = seemingly unrelated regression; B = regression coefficient; SE = standard error; PHQ-9 = depression measure; GAD-7 =
anxiety measure; LTC = long term medical condition; SE = standard error; COPD = chronic obstructive pulmonary disease; the reference
category in this analysis = no self-reported long term condition; all coefficients controlled for demographic and treatment-related variables
as for the main analysis (table 3); * p <.05; ** p <.01; *** p <.001

Additional references

34. Muraki E. A generalized partial credit model. In Handbook of modern item response theory (eds
WJ van der Linden, RK Hambleton): 153-164. New York: Springer, 1997.

35. Teresi JA, Ocepek-Welikson K, Kleinman M, Eimicke JP, Crane PK, Jones RN, Lai JS, Choi SW,
Hays RD, Reeve BB, Reise SP, Pilkonis PA, Cella D. Analysis of differential item functioning in the
depression item bank from the Patient Reported Outcome Measurement Information System
(PROMIS): An item response theory approach. Psychol Sci Q 2009; 51:148-180.
Impact of long-term medical conditions on the outcomes of
psychological therapy for depression and anxiety
Jaime Delgadillo, Alexander Dawson, Simon Gilbody and Jan R. Bhnke
BJP 2017, 210:47-53.
Access the most recent version at DOI: 10.1192/bjp.bp.116.189027

Supplementary Supplementary material can be found at:

Material http://bjp.rcpsych.org/content/suppl/2016/11/09/bjp.bp.116.189027.DC1

References This article cites 25 articles, 4 of which you can access for free at:
http://bjp.rcpsych.org/content/210/1/47#BIBL
Reprints/ To obtain reprints or permission to reproduce material from this paper, please
permissions write to permissions@rcpsych.ac.uk

You can respond /letters/submit/bjprcpsych;210/1/47

to this article at
Downloaded http://bjp.rcpsych.org/ on December 1, 2017
from Published by The Royal College of Psychiatrists

To subscribe to The British Journal of Psychiatry go to:

http://bjp.rcpsych.org/site/subscriptions/