Introduction
Policymakers and governments must decide their healthcare priorities on the basis
of the best healthcare intelligence available to them. Recent interest has increasingly
focused on the global implications of an increasing and elderly population with long-term
conditions.
1–3
The most recent figures from the Global Burden of Disease Study 2010 show that the
third top global cause of death was chronic obstructive pulmonary disease (COPD),
4
rising from fourth place in 1990.
5
It is predominantly caused by cigarette smoking and leads to lung airflow limitation,
cough, excessive sputum production, and breathlessness. People with COPD can suffer
from substantial disability as the condition progresses.
6
A pressing challenge for governments is how best to project the future trend in the
prevalence and burden of COPD in order to plan adequate health and social care for
those affected by this condition within the scope of limited resources. Governments
should ideally be planning for COPD on two levels: (1) they should consider how to
manage resources to care and treat people who are already affected by COPD; and (2)
how to prevent a greater increase in the burden from COPD by minimising the continuing
smoking epidemic.
In order to make such calculations, governments and other healthcare providers need
to draw on epidemiological models. Merriam-Webster's dictionary defines a ‘model’
as ‘a system of postulates, data, and inferences presented as a mathematical description
of an entity or state of affairs’. This is a useful starting point when considering
the role of models in epidemiology. Most models are explanatory in nature and describe
the relationships between different parameters. The focus of this study is on models
which help to project future epidemiological trends and patterns in populations with
COPD. Governments and policymakers have access to many models, but a review is required
to appraise the published COPD models to aid selection between them.
Various features of COPD present a particular challenge to mathematical and epidemiological
modelling, including the many different definitions of a COPD diagnosis and its overlap
with a diagnosis of asthma. Although COPD is most clearly attributable to cigarette
smoking, there is debate over how best to classify non-smokers who develop COPD with
the immunological and pathological features of COPD as a result of exposure to occupational
dusts and gases or recurrent chest infections. In addition, there is uncertainty as
to the correct classification of older non-smoking adults who have evidence of lung
cell remodelling including squamous metaplasia following chronic inflammation due
to long-term asthma. Such older adults have often lost the reversibility in their
airways obstruction and demonstrate spirometry which is consistent with the thresholds
for COPD.
7–9
According to the Global Initiative for Chronic Obstructive Lung Disease (GOLD), the
diagnosis of COPD is characterised by an obstructive lung defect with forced expiratory
volume in one second to forced vital capacity (FEV1/FVC) ratio <0.7.
10
Controversy regarding this threshold also complicates decisions of precisely which
population to include in modelling. Lung function decreases with age, so a proportion
of elderly people (age 75+) who have never smoked still fit these criteria for COPD.
Some doctors reasonably argue that such elderly people really have normal lung function
for their age and that medicalisation of the elderly should be avoided.
7
An alternative threshold of the lower limit of normal for FEV1/FVC has been proposed
with a decreasing threshold according to age by percentile. The bottom 5% of FEV1/FVC
measurements for whichever total population being measured would be considered abnormal
in the older age group.
9
However, no up-to-date large standardised population database currently exists to
validate such a measure. The nearest is the use of the European Coal and Steel Workers
Population to provide percent predicted FEV1 values; however, this population was
standardised over 20 years ago and is based on a working white European population
without ethnic minorities.
11–13
Similarly, younger people (age 30–40 years) with larger FVC values and greater respiratory
reserve may already have sustained COPD-type damage to their lungs before they reach
the <0.7 ratio threshold, so at this end of the age range there is a risk of under-diagnosis
of COPD.
13
The debate regarding the diagnosis of COPD is more than just a debate over spirometry
thresholds. As many developing countries do not have access to spirometry or even
to a reliable power supply, the usefulness of such diagnostic thresholds is limited.
It has been proposed that COPD may also be diagnosed on history and clinical features.
However, studies have shown that using clinical indicators of pulmonary function to
diagnose COPD missed many participants who had low lung function and airways obstruction,
especially in current smokers.
14
Therefore, in many countries the current situation has evolved where COPD is diagnosed
from physician opinion without corroborating evidence from spirometry, resulting in
a significant overlap between a diagnosis of COPD and a diagnosis of asthma.
It seems likely that classifications in the future will evolve as the role of host
susceptibility is increasingly understood in terms of genetic and epigenetic features.
Several candidate genes related to COPD have been identified.
15
In addition, the science of epigenetics helps to explain how DNA transcription has
been activated or suppressed by DNA methylation, acetylation, or other mechanisms
in response to predominantly prenatal and early life environmental influences.
16
The result of such switching on or off of DNA transcription is to determine the host's
response to noxious stimuli including cigarette smoke. Increased understanding of
these factors is helping to unravel the mysteries of why some life-long smokers are
virtually unaffected by their habit while others have severe COPD. Estimates as to
the prevalence of COPD among smokers aged >45 years vary from 15% to 50% according
to the criteria used for diagnosis.
17,18
Modelling COPD is also challenged by the key feature of exacerbations. An exacerbation
may be triggered by increased bacterial or viral load in the lungs which induce an
aggressive immune response and associated clinical features.
19–21
Associated with a greater frequency of exacerbations is higher morbidity, due to faster
disease progression in terms of loss of lung function, and also mortality.
21
An additional challenge is the level of mathematical sophistication within each model.
Ideally, a researcher with considerable statistical skill would be available to check
the algorithms that drive each model and so provide a full appraisal of the quality
of each model. In the absence of this ideal, it was decided to appraise the quality
of reporting of each model as a proxy for the model's mathematical quality. Taking
these challenges into account, it will be necessary to describe a degree of context
with each model in order that it can be applied in an appropriate setting. This will
help subsequent researchers to understand the necessary caveats to include when describing
the results from each model.
Objectives
To identify all available models for estimating projections of COPD prevalence and
burden, and to assess the quality of reporting of each model in its key publication.
Methods
A search strategy has been developed using search terms to cover the three concepts
of ‘modelling’, ‘disease burden’, and ‘chronic obstructive pulmonary disease’ (see
Appendix 1 for full details). Searches will be conducted in the following electronic
databases: MEDLINE, EMBASE, CAB Abstracts, World Health Organization (WHO) Library
and Information Services (WHOLIS — library catalogue of books and reports), WHO Regional
Indexes (AIM (AFRO), LILACS (AMRO/PAHO), IMEMR (EMRO), IMSEAR (SEARO), WPRIM (WPRO)),
and a modified search strategy will be used to identify reports from the WHO home
website and Google. Searches will be for both published and unpublished modelling
studies from 1980 (when modelling methods first began to be widely used) to 2013.
Two authors will independently review the studies against the inclusion criteria and
make a decision as to whether the study is suitable. Disagreements will be resolved
by discussion and, if this is not possible, a third reviewer will arbitrate.
Inclusion criteria
Any modelling study which uses demographic and epidemiological data to project the
prevalence and disease burden will be included. The included projected outcomes which
are of interest are one or more of: incidence, prevalence and mortality, and disease
burden. With regard to ‘disease burden’, the outcomes of interest can be considered
from the individual's point of view, from the point of view of the healthcare system,
and from the point of view of broader society. For the purposes of this review, the
focus is on the perspective of the healthcare system. Other perspectives are valid;
however, different instruments are used to measure them and the purpose of this study
is to guide policymakers who will focus on the healthcare system perspective. Quality-adjusted
life years (QALYs) and disability-adjusted life years (DALYs) are often used to measure
and quantify the burden to the individual of the morbidity they are suffering. Treatments
are assigned a cost per restored QALY, and this is an important measure used in cost-effectiveness
studies. However, the scope of this study is more limited in order to avoid confusion
of perspectives. Some of the studies included may discuss QALYs and DALYs, but they
have not been chosen as primary disease burden outcomes for this review. Instead,
we will concentrate on primary care visits, emergency department visits, hospital
admissions, and COPD treatment costs.
Exclusion criteria
There will be no exclusions on the basis of language of the report. Studies which
are population-based surveys of prevalence without modelling will be excluded as there
has recently been a systematic review of such studies.
22
‘Models’ will be excluded if they describe animals, cell lines, clinical series, or
estimates of individual risk (such as individual prognostic models). Decision analytical
models or decision support models will be excluded where they refer to clinical decision-making
for individuals rather than populations. Models that compare one intervention with
another intervention will also be excluded, as the aim is accurately to project the
baseline outcomes so it is premature to take into account the effect of interventions.
Also excluded will be regression models which start with a COPD population and ‘back-calculate’
the prevalence or burden using regression to quantify risk factors, as this follows
a different logic from that of projection modelling.
Participants
The source population for the model may be from anywhere in the world. The model will
pertain to adult populations aged >40 years as it is usually not appropriate to diagnose
COPD in younger people.
10
COPD may be diagnosed by physician, spirometry, or by questionnaire. Other assumptions
regarding the diagnosis of COPD will be evaluated in the context of the model.
Data extraction
The data will be extracted by one author and checked by a second. Data will be extracted
using a pre-piloted data extraction form. The following identification details will
be extracted for each model: author and email address, year, institution, and funding
source. These data will be followed by: the purpose of the model, model title, model
type, model setting, time period, and population (age, sex and country). Also extracted
will be: inputs to the model, source of input data, details of processing of the model,
outcomes for COPD (incidence, prevalence, mortality, GP visits, emergency department
visits, hospitalisations, treatment costs), model output/results, details of the model's
availability, any comparisons with other studies, social and economic policy implications
of model outcomes, and future research recommendations. In this way, the data extraction
form aims to encompass a comprehensive picture of the model.
Quality appraisal framework
Ideally, a quality appraisal of the actual modelling process would be undertaken.
However, this requires significant statistical technical expertise. A pragmatic decision
has therefore been made to quality appraise the reporting of the models rather than
the actual modelling process for those that have full published reports. In order
to do this, a quality of reporting framework has been designed following review of
key guidelines as to good practice in modelling.
23–26
A scoring mechanism was devised in collaboration with Simon Capewell of Liverpool
University
27
to weight the importance of the different elements required to produce a relevant
high-quality model (see Appendix 2).
Strategy for data synthesis
The study will be the unit of analysis. Models will be described and classified. A
detailed critical narrative synthesis of the highest scoring models will be undertaken.
Where the models are not available, we will write to the model authors for further
clarification. No subgroup analysis is planned.