Psoriasis and the risk of diabetes: A prospective

population-based cohort study

Background: Data evaluating the impact of objectively measured psoriasis severity on type 2 diabetes
mellitus (T2DM) risk are lacking.

Objective: To determine the risk for T2DM in patients with psoriasis compared with that in adults without psoriasis,
stratified by categories of directly assessed body surface area (BSA) affected by psoriasis.

Methods: A prospective, population-based, cohort study from the United Kingdom in which 8124 adults with
psoriasis and 76,599 adults without psoriasis were followed prospectively for approximately 4 years.

Results: There were 280 incident cases of diabetes in the psoriasis group (3.44%) and 1867 incident cases of
diabetes in those without psoriasis (2.44%). After adjustment for age, sex and body mass index, the hazard ratios
for development of incident diabetes were 1.21 (95% confidence interval [CI], 1.01-1.44), 1.01 (95% CI, 0.81-
1.26), and 1.64 (95% CI, 1.23-2.18) in the groups with 2% or less of their BSA affected, 3% to 10% of their BSA
affected, and 10% or more of their BSA affected compared with in the groups without psoriasis, respectively (P =
.004 for trend). Worldwide, we estimate an additional 125,650 new diagnoses of T2DM per year in patients with
psoriasis as compared with in those without psoriasis.

Limitations: Relatively short-term follow-up and exclusion of prevalence cases, which may have masked
associations in patients with less extensive psoriasis.

Conclusion: Clinicians may measure BSA affected by psoriasis to target diabetes prevention efforts for
patients with psoriasis.

Key Words: body surface area; cohort study; diabetes; epidemiology; psoriasis.

P soriasis is a common, chronic inflammatory disease affecting more than 125 million peo-ple worldwide.1,2 A vast
and growing body of literature links psoriasis to an excess risk for major medical morbidities, and even
mortality.3-10 Of spe-cial interest is the association of psoriasis and diabetes,11,12 as genetic studies have
identified shared susceptibility loci and inflammatory cytokines that are up-regulated in psoriasis and
known to promote insulin resistance.13-17 A recent, small meta-analysis identifying 5 articles (4 of which
did not measure psoriasis severity) assessing incidence showed that psoriasis is associated with a
relative risk of 1.27 (95% confidence interval [CI], 1.16-1.40) for development of type 2 diabetesmellitus
(T2DM).4 Beyond the systematic review, we identified 1 additional study that used treatment patterns to
establish a higher risk for T2DM.18 Despite current knowledge, data from population-based, prospective
studies evaluating the impact of objectively measured disease severity on T2DM risk are still lacking.
Only a handful of studies have assessed risk (incidence) of T2DM while adjusting for confounding
variables,11,18-20 and even fewer have been able to evaluate the effects of psoriasis severity (all of which
used treatment patterns) on T2DM association.18,19
Therefore, a better understanding of how physician-reported psoriasis severity affects the risk for
development of T2DM is essential to determine which patients have a clinically important increased risk
for diabetes and therefore should be targeted for augmented prevention efforts. Although a variety of
approaches have been used to define psoriasis severity, categories of body surface area (BSA) affected
have been commonly used in epidemiologic studies and are clinically intuitive. The National Psoriasis
Foundation21,22 and the Centers for Disease Control23 categorize extent of BSA affected into 2% or less
(mild), 3% to 10% (moderate), and more than 10% (severe). Furthermore, we have previously shown
that these simple categories are positively associated with prevalence of psoriatic arthritis and major
medical comorbidities in a doseresponse manner.12,24 The goal of this study was to determine the risk
for development of T2DM in patients with psoriasis compared with in adults without psoriasis, stratified
by categories of directly assessed BSA affected by psoriasis.

METHODS
Study design and data source
We conducted a prospective, population-based cohort study nested within The Health Improvement
Network (THIN) to determine the incidence of diabetes in patients with psoriasis. THIN is a United
Kingdom (UK) electronic medical record database containing patient demographics, diagnostic
information, clinical measurements, and prescriptions from general practitioners (GPs) using Vision
software (In Practice Systems, Ltd, London, UK). GPs coordinate almost all patients’ care in the UK
health care system; hence, medical information from specialists and hospitals is routinely recorded in
patients’ electronic medical record. Diagnoses are recorded in THIN through a READ diagnostic code
system,25 and documented prescriptions are linked to the British National Formulary. 26 This study’s
version of THIN included longitudinal data on 7.5 million registered patients from 415 practices, with
demographics broadly representative of the general UK population.27 Studies have validated the
accuracy of THIN data for use in large-scale epidemiology research, particularly in psoriasis
research.28,29 This study was conducted in compliance with the Declaration of Helsinki, 30 and the
manuscript was prepared in accordance with the Strengthening the Reporting of Observational Studies
in Epidemiology statement.31 The research was approved by the University of Pennsylvania Institutional
Review Board and the Cambridgeshire Research Ethics Committee. No consent was required, as
deidentified data were used.

Study population
The Incident Heath Outcomes and Psoriasis Events cohort, which was previously described by our
group,12 comprises individuals randomly sampled from THIN who, at the time of sampling, were
between the ages of 25 to 64 years, had at least 1 psoriasis READ diagnostic code within 2 years before
survey administration, and were registered in a practice that was participating in THIN’s Additional
Information Services (58% of the THIN practices), which involves participants’ GPs completing
questionnaires in exchange for financial compensation. As psoriasis was the exposure of interest, GPs
were prospectively sent a research survey, 29 which was collected in the subsequent 12 months and
involved GPs confirming psoriasis diagnosis and evaluating severity of disease. GPs assessed the
extent of psoriasis by evaluating the amount of BSA involved and provided this as a continuous estimate
(0%-100%) and as separate categories as follows: mild (limited disease with #2% of BSA affected),
moderate (scattered disease with 3%-10% of BSA affected), or severe (extensive disease with [10% of
BSA affected).32 Each patient with psoriasis was individually matched with up to 10 randomly selected
patients without psoriasis (defined by no history of psoriasis READ diagnostic codes) by practice, visit
date, and age (within 10-year age categories). The exposed group in our primary analyses required GP
confirmation of psoriasis diagnosis and severity assessment by BSA affected from the survey. The
unexposed group was also required to be actively registered, with at least 1 visit to their GP within 2
years before the time of random sampling. Subjects with and without psoriasis were excluded from
analyses if there was a history of T2DM at baseline, which corresponds to the index date.

T2DM outcome and covariate definitions

The primary outcome, namely, the incidence of T2DM, was defined by a composite of 1 T2DM READ
diagnostic code plus a second READ diagnostic code, 1 T2DM pharmacologic therapy, or 1 laboratory
confirmation, whichever occurred first. This identification approach was adapted from a validation study
using THIN by Sharma et al with 100% PPV.33 Furthermore, according to the UK National Institute for
Clinical Excellence and GP Notebook clinical guidelines, T2DM was defined as a glycosylated
hemoglobin level of 6.5% or higher or 48 mmol/mol or higher. 34,35 T2DM antidiabetic drugs were
identified according to the British National Formulary that corresponded with the observation period.36
Multiple potential confounders were identified a priori: age, sex, body mass index (BMI), smoking and
alcohol use, medical comorbidities (hyperlipidemia and hypertension), use of prescription oral and
inhaled corticosteroids, and the Townsend deprivation score (which correlates with socioeconomic
status). These were measured before survey sampling or the corresponding visit date for matched
controls. Age was treated as a continuous variable, whereas the other covariates were categorical.

Person-time calculation
For psoriasis-exposed individuals, the index date (cohort entry) was defined by the date the GP survey29
was sent. The index date for unexposed individuals was the corresponding date for the exposed
individual to whom they were matched. The event of interest was the incidence of T2DM. Patients were
censored at end of follow-up as a result of patient death or transfer out of practice, end of practice
participation in THIN, or the end of the observation period (February 5, 2015).

Statistical analysis
Baseline patient demographics were summarized with descriptive statistics for patients with and without
psoriasis. Differences in baseline demographic and clinical characteristics were assessed using the
Student t test for continuous variables and x2 tests for categorical variables. A raw incidence rate for
T2DM was estimated by dividing the number of events (new diagnoses of T2DM) by the total persontime
under observation. Cox proportional hazards models were used to estimate hazard ratios (HRs)
comparing the exposed and unexposed cohorts. We identified covariates for inclusion in a final,
adjusted Cox proportional hazards model utilizing a purposeful selection approach. Our initial model
included all covariates in Table I. Variables were then eliminated from the model by sequentially
removing covariates, starting with the largest P value (P[.05). If the HR point estimates for psoriasis did
not change by more than 10% in the reduced model as compared with in the prior multivariable model,
then the covariate was removed. Our final model was adjusted for age, sex, and BMI. The likelihood
ratio test was used to evaluate significance of covariates, determining best model fit. Log-log survival
plots for psoriasis were constructed to assess the proportional hazards assumption, which was not
violated. All effect measures were reported with 95% CIs. Our sample size provided power in excess
of 98% for HRs of 1.2 or greater, assuming a 2-sided a level of 0.05. Analyses were performed with
STATA 13.1 software (StatCorp LP, College Station, TX).

RESULTS
Patient characteristics
Of the 10,474 eligible patients with psoriasis READ codes sampled for survey mailing, 10,026 had
surveys returned from their GPs, of whom 9069 had a confirmed diagnosis of psoriasis by their GP. A
total of 652 subjects had a history of T2DM, and 17 had missing medical records and were therefore
excluded from analyses. Our entire study cohort consisted of 84,723 individuals of whom 8124 had
psoriasis and 76599 did not. After stratification by physician-reported BSA, 4216 (51.90%), 2915
(35.88%), and 993 (12.22%) had 2% or less of their BSA affected, 3% to 10% of their BSA affected,
and more than 10% of their BSA affected, respectively (Fig 1). Compared with patients without
psoriasis, those in the group with psoriasis had the same median age, slightly more male patients, and
similar follow-up duration. Standardized differences (Table I), a measure not influenced by sample size,
indicated meaningful differences ([0.1) between patients with and without psoriasis
for BMI and smoking status.37
Incidence of T2DM in patients with psoriasis
There were 280 (3.44%) and 1867 (2.44%) incident outcomes of T2DM in patients with and without
psoriasis, respectively. The unadjusted incidence rate of T2DM was 5.97 new diagnoses of T2DM per
1000 person-years in the unexposed group. The highest incidence rate of T2DM was in patients with
more than 10% of their BSA affected by psoriasis (12.22 per 1000 person-years) (Table II). After
adjustment for age, sex, and BMI, the HRs for development of incident T2DM were 1.21 (95% CI, 1.01-
1.44), 1.01 (95% CI, 0.81-1.26), and 1.64 (95% CI, 1.23-2.18) in the groups with 2% or less of their BSA
affected, 3% to 10% of their BSA affected, and more than 10% of their BSA affected compared with in
patients without psoriasis, respectively (P = .004 for trend; Table III). As patients with more than 10%
of their BSA affected by psoriasis may be heterogeneous in their disease burden, we further evaluated
increases in BSA affected in this group (on the basis of a continuous measure provided by GPs) and
determined that for every 10% increase in BSA affected with the fully adjusted model, the hazard of
incident T2DM increased by a factor of 1.193 (95% CI, 1.025-1.390). Interactions between age and
psoriasis and sex and psoriasis were not statistically significant (P \ .1). Findings were similar when
robust standard errors were used to account for clustering within matched sets of observations.

Sensitivity analyses and multiple imputation

Our observations were robust across a variety of sensitivity analyses and after use of multiple
imputation for missing BMI data (Supplemental Table I; available at http://www.jaad.org). The results
changed little in response to varying outcome (T2DM) definitions, varying definitions of exposure
(psoriasis) group, and alternative handling of missing data in BMI. Investigating observation bias and
treatment pattern effects did not affect our findings significantly (Supplemental Table I, lines 11, 7, and
8). To ensure that incident T2DM outcomes were not prevalent outcomes, sensitivity analyses excluding
patients who had registered with the practice within 9 months before the index date38 and those with
diagnoses of T2DM within 1 year after index date were conducted, and the results were essentially
unchanged.

Attributable risk
On the basis of our data, we estimate that in individuals with psoriasis, there are 3 additional diagnoses
of diabetes per 1000 patient-years compared with in patients without psoriasis, after accounting for
traditional risk factors such as age, sex, and BMI identified in routine medical practice. Furthermore, the
excess risk, or number of extra cases of T2DM, attributable to psoriasis is most clinically significant in
patients with more than 10% of their BSA affected, with 6.25 extra cases per 1000 person-years as
compared with in those without psoriasis versus in those with 2% or less of their BSA affected, in which
case the excess risk is 1.99 per 1000 person-years. When standardized cumulative incidence estimates
based on our adjusted Cox model39 are used, an additional estimated 125,650 new diagnoses of T2DM
per year worldwide are seen in patients with psoriasis as compared with in those without psoriasis, of
whom approximately 25,000 have psoriasis affecting more than 10% of their BSA.

DISCUSSION
In our large, population-based, prospective cohort study from the United Kingdom, we found a positive
dose-response relationship between an objective measure of psoriasis severity and the incidence of
T2DM. Patients with more than 10% of their BSA affected by psoriasis had the highest risk for
development of T2DM compared with that in adults without psoriasis. Moreover, for every 10% increase
in BSA affected by psoriasis (ie, 20%, 30%, etc) there is approximately a 20% higher hazard of diabetes,
which emphasizes the dose-response relationship between burden of skin psoriasis and risk for
development of diabetes. Our results accounted for major diabetic risk factors (age, sex, and BMI)
routinely collected in clinical practice, were robust across multiple sensitivity analyses, and suggest that
psoriasis and its severity are important predictors of diabetes risk. We estimate, on the basis of our
data, that patients with psoriasis affecting 10% or more of their BSA have about a 60% higher risk per
year for development of T2DM, which translates into an extra 25,000 new cases of diabetes annually
worldwide that are attributable to severe psoriasis. We also observed an increased risk for diabetes in
patients with 2% or less of their BSA affected by psoriasis, and although this association was smaller
and of lower clinical significance, it was still important to note. Interestingly, we did not observe a
statistically significantly increased risk in the group with 3% to 10% of their BSA affected. We note that
on the basis of the 95% CIs and Kaplan-Meier curves (Fig 2), we cannot exclude that this finding is
statistically different from that for the group with 2% or less of their BSA affected by psoriasis, although
the risk is evidently lower than that for the group with more than 10% of their BSA affected. Furthermore,
we have previously demonstrated a dose-response relationship between BSA and prevalent diabetes,12
so it is conceivable that our findings with incident T2DM were diluted by prevalent T2DM in patients
with varying levels of psoriasis severity. Additionally, the sample size and duration of follow-up may not
have been sufficient to detect a prospective association in this subgroup.

Our study advances existing literature by demonstrating that a simple phenotypic objective measure of
disease severity has clinical significance beyond the skin in that it predicts risk for development of
T2DM. It also builds on our recent work in which we demonstrated that mortality risk among patients
with psoriasis followed prospectively for a period of approximately 4 years is restricted to patients with
more than 10% of their BSA affected.10 Current literature uses treatment patterns (ie, systemic therapy)
to define severity, which is problematic, as treatment may alter the risk for diabetes (either positively or
negatively), and it is well known that psoriasis is widely undertreated. 21 Thus, it is necessary to
determine whether the findings generalize to patients with extensive psoriasis who are not receiving
systemic agents or phototherapy. Most studies that assessed incidence 4,18,19 did not adjust for
confounders, and all used psoriasis treatment as a surrogate marker for psoriasis severity. Moreover,
our broadly representative, population-based study design and high GP survey response rate suggest
that our findings are likely generalizable to the psoriasis population (ie, that they have strong external
validity).

Overall limitations of our study include selection bias and information bias. However, exposed and
unexposed matched individuals were randomly drawn from the same population (minimizing selection
bias), and data were collected by the same practices within a similar time frame (minimizing information
bias). It is possible that patients with severe psoriasis visit their GP more frequently, especially with
existing comorbidities, although our findings were robust to observation bias (Supplemental Table 1,
line 11). Finally, as with all observational studies, the possibility of nmeasured or unknown confounders
exist. However, we tested and accounted for numerous confounders in our primary analyses, and our
findings were robust to multiple sensitivity analyses.

CONCLUSION
Our findings demonstrate that psoriasis is a significant risk factor for incident T2DM beyond age, sex,
and BMI and that the risk for development of T2DM increases with increasing BSA affected. Clinicians
may consider measuring BSA affected by psoriasis as part of the standard of care because it has
important prognostic implications. Patients with psoriasis affecting more than 10% of their BSA should
be targeted for diabetes prevention efforts.