Protocol registration

The protocol for this review was drafted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement and the Meta-analyses of Observational Studies in Epidemiology (MOOSE) reporting guidelines. It was registered on PROSPERO and is available from http://www.crd.york.ac.uk/PROSPERO/display_record.php?ID=CRD42018087066, see online supplementary file 1.

Search strategy and inclusion criteria

We conducted electronic searches of MEDLINE, EMBASE and Cumulative Index to Nursing and Allied Health Literature  (CINAHL) databases from inception through to February 2019, using the search strategy specified in online supplementary file 2. This was developed with an information specialist. Electronic searches were complemented by manual searching through reference lists of studies that were identified for potential inclusion as well as backward and forward searching. We also searched the WHO Clinical Trials registry and ClinicalTrials.gov registries using the key words “normal ageing”, “lung function decline”, “FEV1 decline”, “FVC decline” and “lung decline”.

We included cohort studies and also planned to include the control arms of randomised controlled trials that measured the decline of lung function in an ageing population. The inclusion criteria were:

• Longitudinal studies that followed some or all of the adult participants past the age of 65 years.

• Three or more measurements of lung function undertaken.

• Studies with a follow-up period of 3 years or longer.

• Studies that measure lung function (ie, FEV₁, FVC, peak expiratory flow rate (PEFR)).

We excluded studies if the participants did not meet the prespecified age criteria; if the population of interest were reported to include smokers or those with risk factors such as occupational inhalation; if participants were reported to have respiratory symptoms such as wheeze, dyspnoea or chronic cough or if the study included participants with known respiratory disease such as asthma or COPD.

Study selection and data extraction

Two authors (ETT, MG) independently screened the titles and abstracts of studies identified in the initial search for eligibility. Prior to commencing screening, a small subset of 50 titles were screened by the two reviewers as a calibration exercise to check for >80% agreement. Similarly, after screening, a calibration exercise was conducted for screening the full texts of the studies and targeting >80% agreement. The remaining full texts were retrieved and reviewed independently by the authors to determine eligibility for inclusion. Non-English publications were translated using Google Translate or with the assistance of a translator. Disagreements were resolved by consensus through discussion or with a third reviewer (PG). If there were multiple reports of the same study, the most recent publication with longest length of follow-up was selected for inclusion, and if the two studies had a similar length of follow-up then the study with the largest sample size was included. Two authors independently extracted data from the studies. The Excel data extraction form was piloted using 10 studies prior to data extraction as a calibration exercise to check for adequate agreement (>80%) between the reviewers. Any disagreements were resolved by consensus or with a third reviewer. Extracted measures included study setting, year and duration, participant eligibility criteria, sample size, participants demographics (ethnicity, gender, baseline age), any known risk factors or exposures, baseline lung function, lung function measurements, number and frequency of measurements, average length of follow-up and loss to follow-up. We also aimed to report the proportion of the cohort that subsequently developed symptoms or disease during follow-up.

We assessed risk of bias of included studies using the six items of the Newcastle–Ottawa Scale (NOS)19 for assessing quality of included cohort studies. Disagreements were resolved by discussion or a third reviewer.

Assessed factors included:

• Representativeness of the exposed cohort (eg, low risk: random selection; high risk: non-random selection for example, volunteer sampling).

• Ascertainment of exposure—age (eg, low risk: from medical records; high risk: self-reported).

• Demonstration that the outcome of interest was not present at the start of study (eg, low risk: participants were excluded on the basis of demonstrated airflow limitation; high risk: if participants were not screened).

• Assessment of outcome (eg, low risk: spirometry; high risk: subjective measure of lung function).

• Adequate duration of follow-up (eg, low risk: ≥3 years of follow-up; high risk: <3 years of follow-up).

• Adequate follow-up of cohorts (eg, low risk: <20% attrition, loss to follow-up explained; high risk: >20% attrition, unexplained loss to follow-up).

Studies were assessed as good quality if they had low risk of bias in all six domains, moderate quality if they had low risk of bias in four or five domains and low quality if they had low risk of bias for three or fewer domains.

Statistical analysis

For each study cohort, we extracted the annual decline rates for each lung function measure. If these were not reported, we calculated crude decline rates for all reported lung function measure by subtracting the final measure from the initial measure and dividing the result by the duration of follow-up. If these data were not available, we determined crude rates of decline from the graphs provided or contacted the study authors for original data. The data were first analysed descriptively using graphs to determine whether it was appropriate to pool the data. For continuous outcomes, the mean difference (MD) (or standardised MD if studies used different measuring scales) and SD were calculated. The data were reported as an annual decline (unit/year). The overall rates of decline and corresponding 95% CIs were presented in a forest plot. We planned to perform a meta-analysis to pool the estimates of decline.

We presented the data by functional parameter (FEV₁, FEV_0.75, FVC, PEFR) and planned to compare annual decline rates by sex and ethnicity in absolute and relative terms, where data were available. We also extracted and presented age-specific decline rates by decade of age if studies reported these data. We planned to separately analyse the data of those who developed disease during follow-up. We also planned to examine for birth cohort effects if the data were available. Sensitivity analyses were planned for study duration greater than 10 years.

Patient and public involvement

Patients were not involved in the design, data extraction or data analysis of this review.

Study characteristics

From searches of MEDLINE, Embase and CINAHL performed on 12 February 2019, we identified 4331 records. An additional 54 records were identified from clinical trial registries and reference list searches. From these, we retrieved 143 papers for full text review; 115 of these did not meet our selection criteria and a further 12 were removed as duplicates. In total, 16 studies20–35 were included in the systematic review (with one study contributing two data sets29) (figure 1). The studies included 31 099 participants and were conducted between 1959 and 2014 ranging from 5 to 30 years in duration (table 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Study flow diagram showing the process for inclusion of prospective randomised controlled trials (RCTs) and cohort studies for estimating the rate of lung function decline with age.

Overall age-related lung function decline

A meta-analysis was not performed due to substantial heterogeneity across the included studies, and a narrative synthesis was undertaken instead. Thirteen studies reported changes in FEV₁ as an outcome. All studies demonstrated a decline with age, with overall rates of decline from each study ranging from 9.9 to 56.0 mL/year (median 29.2 mL/year). Seven of these studies examined the differences in rates of decline between males and females, showing greater absolute FEV₁ decline in males (median 43.5 mL/year) than in females (median 30.5 mL/year) (table 2, figure 2). Relative rates of FEV₁ decline were calculated for men in eight studies and for women in six studies that reported baseline FEV₁ values. There was no statistically significant difference between men and women’s relative change of FEV₁ from baseline (p=0.7). FEV_0.75 decline was reported in one study.29 This study provided two data sets (follow-up after 15 years, 30 years) provided in table 2.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

The rate of forced expiratory volume in 1 s (FEV₁) decline in 13 study populations, grouped by sex.

Five studies reported changes in FVC, with rate of decline estimates ranging from 14.1 mL/year in the youngest cohort32 (median age 36.2 years) to 65.6 mL/year in the older cohort24 (mean age 73.0 years). In studies that measured both FEV₁ and FVC over time, there was a greater decline in FEV₁ than FVC in two studies, and greater decline in FVC than FEV₁ in three studies. These measures are average estimates across study participants and do not enable calculation of individuals’ FEV₁/FVC ratios. In the one study where individuals’ FEV₁/FVC ratios were reported as an outcome,26 there was a decline by 0.29% per year.

PEFR was reported as an outcome in two studies,27 30 which showed decline rates ranging from −6.6 L/min/year in females to −11.5 L/min/year in males.

Age-specific lung function decline by decade of age

The age-specific rates of FEV₁ change by decade of age were extracted or calculated from three studies.22 23 28 35 In all but one study, estimates of decline increased from the fourth (age 30–40 years) to eighth decades of life (table 3). One study could not be included in this comparative analysis as they included smokers and reported decline rates at the end of study follow-up (rather than baseline age).35 This study reported that the rates of relative decline increase from the seventh (−1.7 %/year) through to the tenth decade (−3.1 %/year), though absolute rates of decline varied. Another study also reported that the rate of decline may be non-linear in multiple regression models of FEV₁ and FVC decline (where age squared was also a statistically significant variable).34

Two studies examined lung function change within age brackets that did not conform to our decade-specific analysis. Bartholomew and Knuiman21 reported greater decline rates in never smokers aged above 45 years (females: −30.7 mL/year, males −45.8 mL/year) compared with those aged below 45 years (females: −24.3 mL/year, males: −36.8 mL/year). Lange et al 25 compared decline rates in both male and female non-smokers in 20-year age groups. Females aged 60–79 years had the greatest decline rates (−31.7±2.1 mL/year) compared with the 40–59 age group (−17.7±1.4 mL/year) and the 20–39 age group which reported an increase of 5.0±2.7 mL/year. Similarly, males aged 60–79 years had the greatest decline rates (−37.1±3.7 mL/year) compared with the 40–59 year age group (−24.2±2.6 mL/year) and the 20–39 year age group (−4.6±4.2 mL/year).

Overall rates of mortality/symptom/disease development

Few studies reported these outcomes in an initially asymptomatic, non-smoking population. One study (Proctor et al)30 reported 85% mortality rate in the elderly cohort (age range 79–96) over 8 years. Another study (Lange et al)25 reported that in their study of non-asthmatics, 364 (2%) patients who did not report having asthma at the beginning of the study, later reported it in follow-up. However, this estimate included smokers. One study (Wang and Petsonk)33 performed their analyses on a highly screened population, meaning they excluded participants from all analyses who developed disease or symptoms during study follow-up. No studies reported the rates of lung function change in those who developed disease during the course of the study compared with those who did not.

Sensitivity analyses

Heterogeneity in study duration was explored in figure 3. After removing studies with a follow-up of less than 10 years, the median rate of decline of FEV₁ was 22.4 mL/year (figure 4).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

The rate of forced expiratory volume in 1 s (FEV₁) decline in 13 study populations by years of follow-up. The size of the circle corresponds to individual study sample size.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Sensitivity analysis, excluding studies with less than 10 years of follow-up. The size of the circle corresponds to individual study sample size. FEV₁, forced expiratory volume in 1 s.

Predictors of the rate of decline in lung function in people without known lung disease

Risk of bias

Risk of bias was determined using a modified version of the NOS19 (figures 5 and 6). No studies received low risk of bias in all domains, but four studies had a low risk of bias in all but one domain.23 28 31 Thirteen studies (81%) were graded as having low risk of bias for representativeness of the population. Six studies (38%) were judged as low risk of bias on how they ascertained the age of the participants (from Medicare eligibility lists or health records). Four cohort studies (25%) clearly demonstrated that pulmonary impairment was not present in participants at the beginning of the study. All studies (100%) used a spirometer to measure lung function which is a validated objective instrument. All studies (100%) had adequate duration of follow-up (3 years or longer). Eight studies (50%) had a high risk of bias for having high attrition rates in their studies (>20%).

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Risk of bias summary for prospective cohort studies estimating the rate of lung function decline with age, assessed using a modified form of the Newcastle–Ottawa Scale. 

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

Graphical representation of the risk of bias in prospective cohort studies estimating the rate of lung function decline with age.

Statement of principal findings

This systematic review of 16 prospective cohort studies conducted in 13 countries provides a summary of all the available evidence looking at lung function change with age. Lung function declines with age in normal, asymptomatic adults with higher rates of decline in absolute lung function parameters in men compared with women. However, the relative rates of decline from baseline between men and women do not differ significantly. The decline in absolute and relative lung function parameters may accelerate with age and is also exacerbated by smoking. We were unable to compare lung function decline rates of different ethnicities due to insufficient data. There was a paucity of longitudinal studies that reported changes in FEV₁/FVC rather than reporting the two parameters separately.

Strengths and weaknesses of the study

This systematic review examined all the available primary studies to allow an examination of the consistency of estimates of decline in FEV₁, FVC, FEV₁/FVC ratio and PEFR. This review particularly focused on older adults; this group is relatively understudied and yet more prone to overdiagnosis and misdiagnosis.6 8 18 While the majority of current prediction equations of lung function are based on cross-sectional studies,36–39 our review searched for longitudinal studies as they change in lung function may provide a complement to measurement at one time point in predicting future lung function.37 Our review included participants who were ageing normally, but may have had non-pulmonary comorbidities such as hypertension and diabetes mellitus. This enabled us to investigate a population that was more representative of a normal ageing population.

Our review has some limitations. We did not have access to unpublished individual participant data to allow the calculation of FEV₁/FVC for the majority of studies, where this were not reported. Five studies separately measured changes in both FEV₁ and FVC, however is difficult to conclude whether the rate of decline in FEV₁ and FVC is proportional. Out of the five studies that reported both FEV₁ and FVC decline, two studies32 demonstrated that FEV₁ declines faster than FVC, but in the three remaining studies,20 21 24 the FVC declined at a faster rate (see table 2). Longitudinal studies that specifically measure the FEV₁/FVC would provide the most reliable measure of this decline. Individual patient data would also allow a more robust analysis of changes in lung function between individuals in the studies.

We were unable to pool the results due to significant heterogeneity across the populations. This review’s findings are also limited by the quality of the included studies, all of which were judged moderate or low quality. Since this review is based on limited populations, the findings may not be generalisable to all individuals, especially those of non-Caucasian ethnicities or from less economically developed countries where smoking and air pollution may be more prevalent for example. The review’s findings may underestimate lung function decline among asymptomatic people, as volunteer bias may be present with cohort studies where healthier individuals may be more likely to participate. Our study aimed to examine the rate of lung function change in the elderly group; however, the majority of included studies did not focus on this age group. COPD misdiagnosis particularly affects those older than 80 years of age; therefore, more studies are required in the elderly group.

Our review did not consider the extent of short-term within-person variation, or ‘noise’, in lung function measurements, which is likely to be considerable.40 41 Any observed change in measurement is a combination of the true change, or ‘signal’, and the random background ‘noise’. The clinical utility of monitoring lung function to decide whether or not COPD is present is in part determined by the ratio of signal to noise in the measurements.42 Changes in measured lung function over a longer period of time may be more likely to indicate some true change rather than just background noise43; therefore, we specified in our inclusion criteria that eligible studies should measure lung function on a minimum of three occasions.

We observed substantial heterogeneity across all the included studies and results. This may be due to inherent differences within the populations studied (including distribution of ages, proportion of men vs women and ethnicities) or the duration of follow-up, or that decline in normal healthy people may vary across individuals without causing disease. We explored differences in duration of follow-up as a potential source of heterogeneity in a sensitivity analysis excluding studies with less than 10 years of follow-up, but found that this did not change the median estimate substantially.

Quality of spirometry, as well as properly maintained and calibrated equipment, causing measurement error and contributing to the ‘noise’ in measurement discussed above, is likely to have contributed to variation in the results. Only nine of the included studies specifically reported that the spirometers used in their studies were calibrated and the measurements had to be acceptable and reproducible, following the American Thoracic Society guidelines on the standardisation of spirometry.44 Two studies used peak flow meters to measure PEFR. These instruments are well known to vary in consistency and accuracy. Maselko et al used a Mini-Wright meter, but the second study by Proctor et al did not specify which peak flow meter they used.

The majority of studies specified that they excluded patients with known disease or symptoms at the commencement of the study. However, most of the studies did not report whether any of the participants in their study sample developed symptoms or respiratory disease in the course of follow-up. Thus, undiagnosed COPD or other respiratory, cardiac, renal or other diseases that cause decline in lung function may have contributed to heterogeneity in the results.

Variation within the results may be also explained by the ‘horse-racing effect’, where an initially low FEV₁ measurement may reflect a greater loss of function in the preceding years and hence predicts faster decline in subsequent years (just as the position of the horse in halfway through the race is related to its speed in the early part of the race and hence speed for the final part of the race).45 46 Regression to the mean, due to inclusion of people with randomly high (or low) measured lung function in the primary studies, may also have contributed to heterogeneity of the results.47 A simple way that primary studies may assess for a horse racing effect, while allowing for regression to the mean, is by constructing Bland–Altman plots of change versus mean FEV₁ level48 (or substituting PEFR for mean FEV₁ as these are highly correlated).49

Comparison with previous research

To date, there have been no systematic reviews or meta-analyses examining the rate of lung function decline with age to assess the potential impact of the fixed threshold on COPD misdiagnosis. Cross-sectional studies have compared people diagnosed with COPD using a fixed threshold and the LLN definition, reporting that the GOLD criteria leads to misdiagnosis of COPD.5–8 50 A prospective cohort study found that the fixed threshold of the GOLD criteria overdiagnosed a large proportion of elderly people over the age of 70 years, and the LLN criteria tended to underdiagnose COPD, when compared with the reference standard which consisted of an expert panel who used all available diagnostic information including spirometry.18

Meaning of the study: possible explanations and implications for clinicians and policy-makers

This review has found that lung function declines with age in all studied populations. The rate of decline appears to accelerate with age, and age-specific estimates of FEV₁, FVC and FEV₁/FVC ratio may be more appropriate for the diagnosis of COPD than the fixed threshold currently used across all ages. Currently, prediction equations for calculating mean lung function values as well as the lower limit of normal (LLN) for all ages are based on data from cross-sectional studies; however; it is argued that this is problematic as they do not factor in the important dimension of time.51 52Spirometers used in practice commonly derive their reference values from the National Health and Nutrition Examination Survey (NHANES), a cross-sectional study which was conducted in the USA between 1988 and 1994. Though the predicted values do reflect a decline in FEV₁ and FEV₁/FVC with age, these decline rates may not be as reliable as the estimates from longitudinal studies included in our review. According to the NHANES III, the median rate of FEV₁ decline for a Caucasian male with an average height of 1.75 m aged between 30 and 80 is 32 mL/year and a female with an average height of 1.6 m has an FEV₁ that declines a median of 25 mL/year. Both of these estimates are lower than the median FEV₁ decline of the studies in our review, which was 43.5 mL/year and 30.5 mL/year for men and women, respectively. Therefore, the predicted age-specific lung function used in spirometers may often mislabel people as having abnormal lung function when they are actually within normal limits.53 More reliable age-specific estimates and prediction equations are required.

Clinicians need to consider whether ‘abnormal’ spirometry results may in fact represent normal ageing. This is especially true for making a formal diagnosis of COPD. If a patient is symptomatic and has airflow obstruction as defined by GOLD criteria, it may be necessary to consider alternative diagnoses such as a dyspnoea of cardiac origin. One proposal for identifying individuals who are experiencing greater loss of lung function than expected is to develop ‘decline charts’ that predict FEV₁ or FEV₁/FVC loss for different ages. This can allow clinicians to monitor lung function over time and assess whether individuals are tracking along expected decline curves. These would also need to account for noise in measurement.

Future research should focus on conducting long-term longitudinal studies in less-studied populations, with emphasis on older adults. These studies should examine the rates of decline in people who eventually become symptomatic or develop disease. This information can guide clinicians to predict what rate of lung function decline may be a prognostic indicator of COPD onset and progression. Further well-designed prospective studies that investigate changes in FEV₁/FVC may allow for the development of algorithms that predict individuals’ expected lung function over time according to their sex, smoking history, age, BMI and ethnicity. The observed change in lung function parameters might then be compared with the expected change to help the clinician determine whether this is extreme enough to warrant diagnosis of disease.