PHIAC 6.7
An Economic Evaluation of Opportunistic Screening For
Chlamydia Trachomatis using a Transmission Dynamic
Model
Pelham Barton and Tracy Roberts, University of Birmingham
1
Status of this Document
This document reports the results of a cost-effectiveness analysis commissioned to inform the
decision making of the Public Health Interventions Advisory Committee of the UK National
Institute for Health and Clinical Excellence. It is based on adapting the model produced for the
Chlamydia Screening Studies (ClaSS). This document was written by Pelham Barton and Tracy
Roberts alone, but we acknowledge input of all of our colleagues and advisors in the ClaSS
project.
Background
The importance of using an appropriate model for the transmission of C. trachomatis to allow for
interaction between individuals has been explained elsewhere (Roberts et al, STI In press). Of the
available dynamic modelling approaches we chose to use discrete event simulation, an individualbased approach, because this was the only approach that would allow the effects of partner
notification to be examined. System dynamics, which uses aggregated data would only permit the
estimation of average effects whereas discrete event simulation allows records of partners
attached to specific individuals to be kept.
We have used the simulation model developed for the ClaSS Study to evaluate the relative cost
effectiveness of opportunistic screening based on new data provided by a recent review of the
clinical literature. In the model, a population is simulated over time, with individual
characteristics changing as necessary on a daily basis. The model was parameterised wherever
possible using empirical data collected in the ClaSS project, with nationally representative data
from studies such as the National Survey of Sexual Attitudes and Lifestyles in 2000 and with new
data from a review of the clinical literature. A substantial improvement of this dynamic model
(and the ClaSS model) compared to others published to date, is the dynamic modelling of the
incidence of long term complications associated with chlamydia. In the original economic
evaluation by Welte et al, the occurrence of these sequelae was modelled statically because the
most cited literature about the complications of chlamydial infection provides the data as a fixed
probability (Stamm et al 1984). We used data from the Uppsala Women’s Cohort Study, in which
the cumulative incidences of ectopic pregnancy and infertility were estimated for a population of
over 40,000 young women in Uppsala County, Sweden (Low et al 2005).
2
Methods
The main features of the model are explained in detail elsewhere (Low et al forthcoming) and are
referred to again only briefly here. The explanation is divided into the following sections: ageing
and replacement; partnership formation and dissolution; chlamydia transmission and progression;
testing and treatment; sequelae associated with pregnancy. The main difference for this report is
the methods applied to the testing and treatment because the current focus is on opportunistic
screening and not population screening as in the ClaSS study. The sequelae associated with
chlamydia include inflammatory complications such as pelvic inflammatory disease in women
and epididymitis in males, sequelae associated with pregnancy such as infertility, ectopic
pregnancy and neonatal complications; conjunctivitis and pneumonia. In modelling terms, the
inflammatory sequelae were modelled as part of the progression of chlamydia, while the sequelae
associated with pregnancy including the neonatal complications, were treated as a separate part of
the model.
Ageing and replacement
The initial population consists of a number of virtual individuals with ages drawn from a uniform
distribution between lower and upper limits. As the model runs, individuals in the model die
(from “other causes”) in line with standard United Kingdom life tables and new individuals at the
minimum age are added to the model.
Partnership formation and dissolution
The initial population does not contain any partnerships and as the model runs, new partnerships
form and are dissolved. Properties of the partnership include frequency of unprotected sexual
contact. Opposite-sex partnerships only are included because of the scarcity of other relavent data.
Individuals were categorised into three sexual activity groups according to rates of sexual partner
change.
Chlamydia transmission and progression
At any time in the model an individual’s chlamydia status is one of the following: no chlamydia;
latent chlamydia; asymptomatic chlamydia; symptomatic chlamydia; inflammation (pelvic
inflammatory disease in women, epididymitis in men).
Testing and treatment
In the absence of a planned opportunistic screening programme, individuals may be treated either
through presenting with symptoms or through background opportunistic screening. For simplicity,
we describe first an opportunistic screening programme aimed only at women. Once the screening
3
programme is under way, any woman within a defined age range has a probability of being
offered screening on any given day. She may or may not accept this offer.
If so, she receives a screening result at a fixed delay after testing. Allowance is made for
sensitivity and specificity of the test used to be below 100%. A woman who screens negative is
not treated, but may be offered screening in the future. A woman testing positive will be treated,
and asked to notify partners. If she complies with partner notification, she will inform current and
former partners within a specified time interval (in the baseline scenario, for simplicity, it was
assumed that notification was either zero or complete). Each partner individually may then
comply with a request to attend for treatment. Any partner attending will be treated without
waiting for a test result.
As well as looking at screening women only, we also considered the possibility of planned
opportunistic screening aimed at men and women.
Compliance is an important issue for both screening methods. The term compliance is used here
as a general term that refers to the uptake rate to the invitation for screening, treatment and partner
notification. In the model compliance with treatment (single dose azithromycin) is assumed to be
100%.
Sequelae associated with pregnancy
Sequelae currently considered in the model are infertility, ectopic pregnancy and neonatal
complications. Instances of infertility are recorded as occasions when a woman does not become
pregnant, but would have done so had her fertility not been reduced by the effect of chlamydiarelated damage.
Calibration of the model
A common problem with modelling, which applies particularly to the ClaSS model, is that the
data available often relate to model outputs rather than model inputs. For example, although the
model requires some individuals to have chlamydia at the start of a model run (input), the steadystate prevalence of chlamydia in the model (output) without population screening is a function of
the transmission dynamics, and the same steady-state prevalence is reached for different starting
patterns of chlamydia. Another example is that the model inputs relate to propensity to form a
new partnership on a given day. In contrast the pattern of partnerships formation which actually
takes place is a model output to be compared with these data.
Thus, instead of direct incorporation of data points as input parameters to the model, it is
necessary to calibrate the model by adjusting the input parameters until a reasonable fit to the
4
available data is obtained. This raises the question of how closely it is desirable to fit the existing
data. The purpose of the model is to estimate the effects over time of introducing a screening
policy, in terms of the difference between the costs and outcomes under different policies
(including a policy of no population screening). Given that sexual behaviour patterns vary over
time, but that it is not feasible to reflect such variation in a model, there is limited value in
producing an extremely close fit to current behaviour patterns. It is more important to test how
robust policy decisions would be to variations in such patterns. It is also worth noting that it is
possible to produce similar observed outputs from a variety of different patterns of input: again, it
is important to test the robustness of conclusions to such variation.
The determination of the input set used to produce the current set of results for the ClaSS model is
described in detail in the following sections. This input set is determined by a process of
calibration. The data shown in Tables 8 to 11 (presented later in this report) represent outcomes
rather than model inputs, and so the model is calibrated to these.
Basic population characteristics
The initial population is evenly distributed in age between 12 and 62, and is assumed to be 50 per
cent female, although this proportion can be altered as a model input. The population is allocated
to activity groups according to sexual partner change in proportions shown in Table 1. Laumann
and Youm use activity groups defined by the number of partners in the past 12 months, calling
those with 0 or 1 partner the ‘periphery’, those with 2 or 3 the ‘adjacent’ and those with 4 or more
the ‘core’(Laumann and Youm 1999). The allocation of activity groups for the ClaSS model is
designed so that the outputs of the model are broadly in line with the data from the lifestyle
questionnaire collected in the ClaSS case-control study but also broadly in accordance with the
Laumann and Youm groups. Thus, these data, presented Table 1 correspond approximately to the
Laumann and Youm activity groups which are reported later. They have been adjusted so the
model outputs fit the data.
Table 1: Initial percentage allocation of sexual activity groups
Male, %
Female, %
Group 1
74
78
Group 2
17
17
Group 3
9
5
Legend:
Group 1 least active, Group 3 most active
In order to ensure that there is some chlamydia in the model, an initial arbitrary age-related
prevalence of chlamydia is applied: zero at age 15, 7% at age 20, 7% at age 25, 2% at age 30 and
zero at age 40. The values of these are unimportant as the steady-state prevalence in the model
5
after the warm-up period depends on the transmission dynamics within the model, not on the
initial state. Each individual is given chlamydia or not, with probabilities appropriate to the
starting age. Linear interpolation is used between the ages given: for example, the probability of a
26 year old starting the model with chlamydia is 6 per cent. Everyone outside the age range 15 to
40 starts chlamydia negative. As the model runs, new (chlamydia negative) 12-year-olds are
added to the model population.
Chlamydia transmission and progression
The main inputs relating to chlamydia transmission and progression are summarised in Table 2.
Except for progression to PID, these parameters were kept from the original models (Kretzschmar
et al 2001, Welte et al 2000). To assess the rate of progression to PID, the model was run for a
high-risk population to replicate the study by Scholes et al (1996). The value used gave a
satisfactory match for the relative risk of PID over a 12 month period.
Table 2: Inputs relating to chlamydia transmission and progression
Parameter
Value
Source
Probability of transmission male to female
0.154
K
Probability of transmission female to male
0.122
K
Incubation period male (days)
10
K
Incubation period female (days)
12
K
Probability asymptomatic female
0.7
K
0.25
K
Recovery rate per day asymptomatic female
0.005
K
Recovery rate per day symptomatic female
0.005
K
Recovery rate per day asymptomatic male
0.025
K
Recovery rate per day symptomatic male
0.03
K
0.0001
W
0.00008
Calibration
Probability asymptomatic male
Progression per day chlamydia to epididymitis
Progression per day chlamydia to PID
a
Legend:
K parameter sustained from Kretzschmar et al. (2001); W estimated from proportion progressing
to epididymitis in Welte R et al. (2000).
a) The progression per day is calibrated to the relative risk reported by Scholes et al (1996).
Testing and treatment
Even in the ‘no screening’ arm, some background screening is assumed. It is assumed that the
highest activity group are more likely to receive background screening. The daily probability of
background screening is age, sex and activity group dependent as shown in Table 3. This is
subject to a minimum gap of 200 days since last screened. These inputs are largely arbitrary but
are informed by the data from the Uppsala Women’s Cohort Study (Table 10). Part of the
6
justification for the model inputs presented in Table 3 is that the Swedish data are applied to a
Swedish population with opportunistic screening only, so differences between the never screened
and screened groups are likely to depend on the level of baseline screening.
Table 3: Assumed daily probability of background screening
Age
Groups 1 and 2
Group 3
Male
Female
Male
Female
15
0.0002
0.0002
0.0004
0.0004
20
0.0004
0.0004
0.0008
0.0008
25
0.0002
0.0002
0.0004
0.0004
30
0.0002
0.0002
0.0004
0.0004
35
0.0001
0.0001
0.0002
0.0002
40
0
0
0
0
Parameters relating to the screening programme are shown in Table 4. The daily probability that
an individual within the appropriate age range is offered screening is calculated so that
approximately 59% of individuals are offered screening in a 12-week period, to match the study
by Senok et al (2005). Individuals are assumed to be consistent compliers or non-compliers. The
minimum period between successive screening was set at 200 days.
Table 4: Parameters for opportunistic screening
Parameter
Value
Source
Daily probability of being offered screening
0.01
Senok
Compliance with screening
0.35
Senok
Waiting time for result of screening (days)
30
Assumption
Sensitivity of screening test (male)
0.999
ClaSS
Specificity of screening test (male)
0.998
ClaSS
Sensitivity of screening test (female)
0.973
ClaSS
Specificity of screening test (female)
0.997
ClaSS
Range in days for recent partner
120
Assumption
Probability that a partner will attend for treatment
0.45
ClaSS
Delay in days for partner to receive treatment
3
Assumption
Legend:
Screening tests are Cobas PCR on urine specimen for men and vulvo-vaginal swab for women
Probability of partner attending for treatment applied independently to each partner
Sequelae associated with pregnancy
With the exception of neonatal complications, all inputs here have been calibrated to the best
available data, using British Office for National Statistics (ONS) data for pregnancy rate and the
Uppsala Women’s Cohort Study for risk of sequelae. Table 5 shows the parameters relating to
7
pregnancy risk, defined as age-related risk of pregnancy per episode of unprotected intercourse.
This is assumed to take into account variation in both fertility and use of non-barrier
contraception. Table 6 shows the other parameters used. Explanation of the risk factors for ectopic
pregnancy and infertility follows the table.
Table 5: Age-related pregnancy risk
Age, years
12.5
Risk of pregnancy
0
per day
17.5
0.0003
22.5
0.0006
27.5
0.0014
32.5
0.0013
37.5
0.0005
42.5
0.0001
45
0.000002
It is generally accepted that repeat infection increases risk of ectopic pregnancy and infertility but
it is not clear whether this is due to separate number of episodes or simply cumulative effects of
the time infected. The model allows either theory to be applied; figures above give a combination
of the two. Consider, for example, a woman who has had two separate episodes of chlamydia
infection, lasting a total of 50 days. With the input parameters above, her risk of ectopic
pregnancy (if she becomes pregnant) is 50 × 0.00003 + 2 × 0.001 = 0.0035. For the same
woman, her fertility is reduced by a factor 1 + 50 × 0.00003 + 2 × 0.0005 = 1.0025. In the model,
an instance of infertility is recorded when a woman who would become pregnant does not become
so as a result of reduced fertility.
Table 6: Other parameters relating to pregnancy
Parameter
Value
Source
0.00003
Calibration
0.001
Calibration
0.00003
Calibration
0.0005
Calibration
0.45
W
Duration (in days) of ectopic pregnancy
50
Advice
Duration (in days) of normal pregnancy
280
Advice
Ectopic pregnancy risk factor (days)
Ectopic pregnancy risk factor (episodes)
Infertility risk factor (days)
Infertility risk factor (episodes)
a
Probability of neonatal complication
Legend:
a) applied to women who are chlamydia positive at the time of giving birth
W, parameter obtained from Welte R et al, 2000.
Costs used in the model
Unit costs used in the model are shown in Table 7.
8
Table 7: Unit costs in the model
Resource use data required
Unit cost, £ Source
Screening tests
21.83 Adams et al (2004)
Treatment of index case including PN
16.53 Adams et al (2004)
Treatment of partners
22.60 ClaSS, PSSRU
a
Infertility
428 NICE guidelines
b
Ectopic pregnancy
2319 HRG costs
PID
2846 HRG costs
Epididymitis
790 Weltec
Neonatal complications
708 HRG costs
Legend:
a) NICE baseline costs for one cycle converted to 2003 costs
b) NICE baseline costs converted to 2003 costs
c) Weighted average of Welte R et al. 2000 converted to 2003 UK costs
Abbreviations: NICE, National Institute for Clinical Excellence; HRG, Health Resource Groups;
PSSRU, Personal and Social Services Research Unit
Calibration of model to existing data
Table 8 shows the observed live births per 1000 women per year in a single run of the model
using each input set, running for 5,000 days after a 5,000 day warm-up period. This is compared
to ONS data for England and Wales in 1998. The fit is reasonable.
Table 8: Live births per 1000 women per year
Age range of mother
Results from model
ONS data
15 to 20
28.91
30.9
20 to 25
75.98
75.5
25 to 30
108.59
102.2
30 to 35
88.30
89.9
35 to 40
37.96
39.8
40 to 45
7.62
7.5
0.18
0.3
a
45 to 50
Legend:
ONS data reported as “45 and over”
Table 9 shows comparisons with ClaSS data applying the ClaSS prevalence survey to the model
at the end of the run. Here the general patterns are approximately preserved.
9
Table 9: Comparisons between model outputs and ClaSS prevalence
study data
Results from model
ClaSS prevalence
study
Mean age difference with partner, years
Age range
Male
Female
Male
Female
16 to 19 y
0.59
-2.41
0.37
-2.21
20 to 24 y
1.65
-2.24
1.41
-2.10
a
Percentage reporting ever having had sex
Age range
Male
Female
Male
Female
16 to 19 y
54.33
80.68
54.2 / 65.1
80.6 / 81.8
20 to 24 y
95.27
99.81
92.1 / 92.5
95.4 / 96.3
Mean length of reported partnership, months
Age range
Male
Female
Male
Female
16 to 19 y
8.78
11.69
7.61
8.44
20 to 24 y
17.64
22.11
17.86
32.24
Laumann and Youm activity groups, percentages
16 to 24 y
Male
Female
Male
Female
Periphery
68.31
70.55
72.6
79.2
Adjacent
20.00
21.29
17.4
15.1
Core
11.69
8.17
10.0
5.7
25 to 39 y
Male
Female
Male
Female
Periphery
85.02
88.98
80.5
91.0
Adjacent
11.76
8.02
12.7
6.6
3.23
3.00
6.8
2.4
Core
Legend:
a) Two sets of data figures are given for percentages reporting ever having had sex. In each
case, the first figure is from the case-control study, the second from the prevalence study.
For long-term sequelae in women, we used data from the Uppsala Women’s Cohort Study. (Low
et al 2005). Information form this was extracted into Kaplan-Meier curves, from which a failure
function could be estimated by age. This represents the probability that an individual will have
experienced a given event at least once by the age stated. This process was replicated by the
model. Table 10 shows the results from the model and from the Swedish database allowing a
comparison to be made. The overall level of sequelae is approximately the same. The good
matching in the never screened groups and the fact that we have much higher results for at least
one positive suggests that our model over estimates the relationship between Chlamydia and
sequelae (as based on the Swedish data) and therefore is likely to lead to relatively more
optimistic results.
10
Table 10: Comparisons of incidence of sequelae associated with
chlamydia in ClaSS project model and Uppsala Women’s Cohort
Study
ClaSS model
Age 25
Uppsala
Age 35
Age 25
Age 35
Infertility
Never screened
0.0032
0.0139
0.0041
0.0308
Always –ve
0.0047
0.0147
0.0044
0.0471
At least one +ve
0.0523
0.1933
0.0052
0.0671
Never screened
0.0033
0.0143
0.0051
0.0187
Always –ve
0.0034
0.0140
0.0038
0.0202
At least one +ve
0.0680
0.2049
0.0053
0.0272
Ectopic pregnancy
Results
The model was run with inputs as described above. The model uses random numbers throughout
and therefore it is necessary to re-run the model a sufficient number of times, with different
random numbers, to ensure the effects of randomisation are reduced. The model was run 26 times,
on a population of 50,000 individuals, with “no screening” (background screening only), for a
total of 10,000 (simulated) days each time. Prevalence of chlamydia was recorded every 20 days
by sex and 5-year age bands.
The baseline results are shown in Figure 1. The vertical line is at 5,000 days and indicates the end
of the “warm-up” period. The part of the graph to the left of the line is the model reaching a
steady state of prevalence. Note that the steady state prevalence was determined by the
transmission dynamics of the model, and not by the initial prevalence incorporated. It can be seen
that the prevalence after the warm-up period remains constant over time.
11
Figure 1: Baseline results of the ClaSS model
The mean prevalence after the warm-up period is shown in Table 11. The prevalence among
females is underestimated compared to the ClaSS survey.
Table 11: Prevalence from model compared with ClaSS survey
Model results, %
ClaSS survey, % (95% CI)
Male
Female
Male
Female
15 to 19
4.27
4.68
3.41 (2.26, 5.15)
6.20 (4.80, 8.59)
20 to 24
5.66
4.35
6.92 (5.22, 8.98)
6.15 (4.93, 8.35)
25 to 29
3.12
1.88
0.62 (0.20, 1.86)
3.27 (2.01, 6.65)
30 to 39
1.34
0.68
0.44 (0.06, 2.94)
0.32 (0.05, 2.34)
a
Legend:
a) ClaSS survey results for age group 16 to 19 years
12
Incorporating opportunistic screening
The model was re-run, introducing opportunistic screening from ages 16 to 24 after the 5,000-day
warm-up period. The results for screening women only and men and women are shown in Figure
2.
Figure 2: Base-case results for screening women only and men and
women
Legend: Top panel women only, bottom panel men and women
13
As in Figure 1, there is a vertical line marking the end of the warm-up period. After the
introduction of planned opportunistic screening, prevalence dropped to a new equilibrium value
particularly in the younger age group where screening is likely to have more effect. The
prevalence dropped slightly more in the case of screening both women and men (Table 12).
Table 12: Steady-state prevalence for baseline run of the model
Age,
years
No screeninga
Male, %
Female, %
Females only
Male, %
Males and females
Female, %
Male, %
Female, %
15 to 19
4.05
4.53
3.83
4.17
3.46
3.75
20 to 24
5.48
4.17
4.94
3.66
4.47
3.36
25 to 29
3.07
1.91
2.78
1.79
2.65
1.71
30 to 39
1.35
0.69
1.29
0.66
1.18
0.66
Legend:
a) Slight variations from Table 11 because of randomness in the model.
Costs and major outcomes averted
In a dynamic model such as the ClaSS model, the results are likely to depend on the time horizon
used for calculations. We took all calculations from the end of the warm-up period, costs and
outcomes being discounted to that point. In line with current guidelines, costs and outcomes were
discounted at 3.5%.
Figure 3 shows the cumulative difference in cost between the three strategies, taken over an
arbitrarily chosen period of up to 12 years from the start of planned opportunistic screening.
14
Figure 3: Cumulative difference in costs
900000
F only v no scr
800000
M and F v no scr
M and F v F only
Incremental Costs
700000
600000
500000
400000
300000
200000
100000
0
0
2
4
6
8
10
12
Time in years
All the figures resulting from the model give three comparisons: screening women versus no
screening, screening men and women versus no screening and screening men and women versus
screening women only. We report outputs from only the first two comparisons because the third is
an artefact generated by the model. Figure 4 shows the difference in the aggregated major
outcomes pelvic inflammatory disease, infertility, ectopic pregnancy and neonatal complications
(given as major outcomes averted). Cumulative costs and outcomes refer only to those incurred up
to the given time. Thus the time horizon is such that we do not consider anything beyond that
point. The model is built on the assumption that a screening programme, once introduced, would
remain in place indefinitely.
15
Figure 4: Cumulative aggregated major outcomes averted
45
F only v no scr
40
M and F v no scr
Major Outcomes Averted
35
M and F v F only
30
25
20
15
10
5
0
0
2
4
6
8
10
12
-5
Time in years
Legend:
Includes cases of pelvic inflammatory disease, ectopic pregnancy, infertility, and neonatal
complications
Cost-effectiveness analysis
Table 13 shows the results of the runs of the ClaSS model, taken to an illustrative (arbitrarily
chosen) period of 12 years after the introduction of planned screening. These calculations give the
estimated results for the first 12 years of a screening programme intended to continue indefinitely.
Because they do not include costs and effects beyond 12 years, they give a slightly conservative
estimate of the costs and effects of a screening programme lasting 12 years only.
Table 13: Summary of results for 12 years follow-up
Results from running individual strategies
Cost (£000)
Major outcomes
No screening
1653
473
Women only
2019
460
Men and women females
2371
441
Comparison between strategies
Difference in
cost (£000)
Major outcomes
averted
ICER
(£/MOA)
F only vs. no screening
366
13
27709
M and F vs. no screening
719
32
22385
Legend:
ICER, incremental cost-effectiveness ratio; MOA, major outcomes averted
16
These results are shown graphically in Figure 5. Here the screening options are shown in
comparison to “no screening”. The incremental cost-effectiveness ratio (ICER) for screening
males in addition to females is lower (more favourable) than that for screening females compared
to no screening. That means that, under the assumptions included in this run of the model, if it is
considered desirable to screen females, then it is more desirable to screen males as well as
females. This phenomenon is known as weak or extended dominance: dominated means that the
option is cheaper and more effective than the comparison. Weak dominance relates to a situation
where two options are both more expensive and more effective than some third option, but the
more effective of the two initial options has a lower ICER than the other two. The option of
screening females only is said to be weakly dominated by the option of screening males as well as
females.
Figure 5: Results over 12 years on the cost-effectiveness plane
800
Men and
women
Additional cost (£000)
700
600
500
Women
only
400
300
200
100
0
0
10
20
30
40
Major outcomes averted
Figure 6 shows the results for a range of time horizons from 6 years to 12 years. The
gradual fall in the incremental cost-effectiveness ratio over time reflects the delay
inherent in a screening programme in which there is a lag before the full effect of the
major outcomes averted as a result of screening become apparent. The values of the
incremental cost-effectiveness ratio were consistently high, suggesting that planned
screening is unlikely to be cost-effective under the conditions built in to this version of
the model. To compare with a threshold of £30,000 per QALY gained, the value for each
major outcome averted would have to equal approximately 1 QALY.
17
Figure 6: Base case results for a range of time horizons
60000
F only v no scr
M and F v no scr
M and F v F only
50000
ICER (£/MOA)
40000
30000
20000
10000
0
6
7
8
9
10
11
12
Time in years
Other scenarios
We considered two other cases. Firstly, we considered opportunistic screening for ages 16 to 29.
Second, as a sensitivity analysis, we considered opportunistic screening for ages 16 to 24
assuming a higher risk of PID, equivalent to that used by Welte et al (2000). Welte et al. used a
probability of 0.25 that an asymptomatically infected woman would develop pelvic inflammatory
disease, and a further conditional probability of 0.4 that this would be symptomatic. Our
definition of pelvic inflammatory disease only included symptomatic cases, this converts to a
probability of 0.1 that an asymptomatic woman would develop pelvic inflammatory disease. For
the ClaSS model this converts to a daily probability of progression of 0.0005 from asymptomatic
chlamydia to pelvic inflammatory disease. In this analysis we used the same probability for
progression from symptomatic chlamydia to pelvic inflammatory disease. Not surprisingly, the
much higher incidence of pelvic inflammatory disease among chlamydia-infected women led to
an increase in the number of major outcomes averted and lower incremental cost-effectiveness
ratios. The results of these sensitivity analysis are summarised in Table 14 and Figures 7 and 8.
18
Table 14: Summary of incremental cost-effectiveness ratios over time
under different conditions
Scenario
Incremental cost-effectiveness ratios,
£/MOA
8 years
12 years
F vs.
none
M&F vs.
none
F vs.
none
M&F vs.
none
Screening ages 16 to 24
25,700
32,200
27,700
22,400
Screening ages 16 to 29
35,300
22,700
23,400
21,000
4,300
5,100
4,400
4,800
Incidence of PID equivalent to Welte
(screening ages 16 to 24)
Legend:
Abbreviations: F, screening women only, M&F, screening men and women, none, no planned
screening; MOA, major outcome averted; ICER, incremental cost-effectiveness ratio
Figure 7: Screening ages 16 to 29
60000
F only v no scr
M and F v no scr
M and F v F only
50000
ICER (£/MOA)
40000
30000
20000
10000
0
6
7
8
9
10
11
12
Time in years
19
Figure 8: Results assuming incidence of PID equivalent to Welte et al
10000
F only v no scr
M and F v no scr
9000
M and F v F only
8000
ICER (£/MOA)
7000
6000
5000
4000
3000
2000
1000
0
6
7
8
9
10
11
12
Time in years
Discussion
The results of the this revised ‘ClaSS’ model suggest that opportunistic screening at estimated
uptake levels of 35%, and assuming a low incidence of chlamydia-associated complications was
not cost-effective. This uptake rate is based on a review of the literature and applied to women
(Senok et al 2005) – which, in the absence of other data, was also applied to men for the purpose
of the current modelling exercise. The model also shows that, provided that the response rate in
men is not much lower than in women, screening men and women is preferred to screening
females only. If the incidence of complications is assumed to be high as per Welte et al (2000),
screening appears relatively more cost-effective.
The strengths of this study are that we used an individual level dynamic mathematical model that
gave the closest approximation to real sexual behaviour in the population together with empirical
data for as many parameters as possible. This model, based on the ClaSS model, is the first to
attempt to incorporate some concept of tubal damage (prevention) caused by repeated or
persistent infection.
There are a number of limitations to these results.
First, these results are based on a single input set. The input set used is determined through a
process of calibration. It is possible to determine further input sets based on different assumptions
of partner mixing and background screening rates, which would necessitate a different calibration
process.
20
Second, due to the long running time of the current model, use of these additional input sets was
beyond the remit of the current study. In terms of the level of uncertainty surrounding the
illustrative cost effectiveness results presented in Table 14, confidence intervals were not
presented. The most appropriate method for exploring uncertainty in these figures is by re-running
the model with another input set. Thus the model needs to be re-run with different input sets to
test the robustness of the conclusions. The model was run 26 times, on a population of 10,000
simulated individuals, with “no screening” (background screening only), for a total of 10,000
(simulated) days each time. The results of calibrating the model were generally consistent with the
empirical data for men but there was a discrepancy between the prevalence expected in the model
under opportunistic screening and that observed in the ClaSS prevalence data for females.
Third, the current calibration, based on new opportunistic screening figures of uptake and relative
risk of PID was difficult for a number of reasons: a compromise had to be found between
calibrating the model both to the prevalence data, sexual activity data and new PID figures. The
calibration process is in itself a lengthy process and producing new results once calibration is
complete takes at least five working days in terms of the running of the model a sufficient number
of times to produce new results. .
The incidence of long term sequelae used in this model were approximated through the calibration
process to be comparable with the results of the Uppsala Women’s Cohort Study. These data
showed a lower incidence than those used in virtually all other cost-effectiveness analyses found
in the literature or which have been used in other studies. In the sensitivity analysis we show that
varying the incidence of PID to concur with the rate typically used in other published studies
produces a lower cost effectiveness ratios. The difference in the results of the sensitivity analysis,
Figure 8, which used an incidence for PID equivalent to that used by Welte et al (2000) compared
to the results of the base case analysis serve to underline the importance of more accurate data on
long term sequelae.
Given the variation in the model, the pattern presented in the graphs are informative but the exact
numbers presented in the summary table should be interpreted with some degree of caution. In
this and the ClaSS model, unlike the analysis by Welte and colleagues, the sequelae are
incorporated into the stochastic model which leads to an increased variance within the model.
This necessitates a proportionally larger overall number of patient runs. The definitions of
sequelae may not match those used elsewhere, making it inappropriate to use data relating to
alternative definitions. In particular, cases of infertility need to be defined very clearly. For the
rarer sequelae such as infertility, the problem is one of definition and to compare with other data,
the definition used within the model is acceptable but not easy to measure in practice. For both
infertility and ectopic pregnancy, it is clinically not clear how these are related to repeat
infections.
21
References
Adams EJ, La Montagne DS, Johnston AR, Pimenta JM, Fenton KA, Edmunds WJ (2004)
Modelling the healthcare costs of an opportunistic chlamydia screening programme.
Sexually Transmitted Infections 80: 363-370.
Kretzschmar M, Welte R, van Den HA, Postma MJ (2001) Comparative Model-based Analysis of
Screening Programs for Chlamydia trachomatis Infections. Am.J.Epidemiol. 153:90-101.
Laumann EO, Youm Y (1999) Racial/ethnic group differences in the prevalence of sexually
transmitted diseases in the United States: a network explanation. Sex Transm Dis 26:25061.
Low N, Harbord RM, Egger M, Sterne JA, Herrmann B. (2005) Screening for chlamydia. Lancet
365:1539.
Low N et al (Forthcoming) Epidemiological, social, diagnostic, and economic evaluation of
population screening for genital chlamydial infection: the Chlamydia Screnning Studies
project.
Roberts TE, Robinson S, Barton P, Bryan S, Low N for the Chlamydia Screening Studies (ClaSS)
Group. (in press) Screening for Chlamydia trachomatis: a systematic review of the
economic evaluations and modelling. Sexually Transmitted Infections
Scholes D, Stergachis A, Heidrich FE, Andrilla H, Holmes KK, Stamm WE. (1996) Prevention of
pelvic inflammatory disease by screening for cervical chlamydial infection. New England
Journal of Medicine 334:1362-6.
Senok A et al. (2005) Can we evaluate population screening strategies in UK general practice? A
pilot randomised controlled trial comparing postal and opportunistic screening for genital
chlamydial infection. Journal of Epidemiology & Community Health 59:198-204.
Stamm WE, Guinan ME, Johnson C, Starcher T, Holmes KK, McCormack WM. (1984) Effect of
treatment regimens for Neisseria gonorrhoeae on simultaneous infection with Chlamydia
trachomatis. New England Journal of Medicine 310:545-9.
Welte R, Kretzschmar M, Leidl R, van Den HA, Jager JC, Postma MJ. (2000) Cost-effectiveness
of screening programs for Chlamydia trachomatis: a population-based dynamic approach.
Sex Transm Dis. 27:518-29.
22