The populations of many developed countries are ageing. In future, therefore, the prevalence of common age-related illness and infirmity during employment is likely to rise. However, there is an economic need to retain skilled and experienced older workers, other considerations allowing. Thus, strategic plans to maximise employment have been announced by the governments of several countries.1

One possible deterrent to full employment at older ages is the potentially greater risk of accidental injury in people taking medication and in those who are limited by sensory, neurological, locomotor, cardiovascular, metabolical, psychiatric or other health impairments. For some kinds of work involving public and third-party risk (eg drivers and pilots) and some health complaints (eg epilepsy), restrictions on employment are legally prescribed.2

The evidence underpinning such restrictions has evolved mostly in relation to driving and the risk of road traffic accidents (RTAs), there being evidence, for example, that defects of peripheral vision,3^–5 glaucoma,4 and use of benzodiazepines6 raise crash and injury rates in drivers. The strength of evidence in other employment situations is less clear but important to establish, as employers need to avoid unjustified restriction of work opportunity while at the same time observing health and safety obligations.

In this paper we report a systematic literature review of chronic health conditions and accident risk that takes as its focus accidents and accidental injuries in the workplace, rather than on the highway or in the home. We have chosen for study a selection of health problems that are common (or likely to become common in an ageing workforce), and could plausibly carry a higher risk of accidents and accidental injury.

METHODS

Search strategy

Systematic searches were conducted of the bibliographic electronic databases MEDLINE (1966 to week 1, November 2006), EMBASE (1980 to week 1, November 2006) and PsycINFO (1985 to week 1, November 2006). The following search terms (medical subject headings and key words) were used:

(1) For work accidents (the outcome): workplace accidents, occupation accidents, work-related accidents, accidents at work, work accidents, accidents occupational, industrial accidents, industrial injury.

(2) For disease and medication categories (exposures of interest):

• Neurological: epilepsy, epileptic, seizure disorder, stroke (cerebrovascular disorder, cerebrovascular disease, cerebrovascular accident (CVA)), transient ischaemic attack, cerebral infarction, Parkinson’s disease, multiple sclerosis, vestibular disease, vestibular disorder, vertigo, labyrinthitis, labyrinth diseases.

• Sensory: visual impairment, vision disorder, cataract, visual acuity, near/far vision defects, reduced field of view, binocular/monocular vision, blindness, partial sight, contrast sensitivity, glaucoma, macular degeneration, diabetic retinopathy, eye disease, hearing impairment, deafness, hearing disorder, hearing loss, sensory impairment.

• Metabolic: diabetes, diabetes mellitus, hypoglycaemia, hypothyroidism, hyperthyroidism, thyrotoxicosis.

• Cardiovascular: dysrhythmia (arrhythmia, heart block), ischaemic heart disease (myocardial infarction, myocardial ischaemia, coronary disease, heart attack, angina pectoris), heart disease, hypertension, syncope.

• Locomotor: arthritis, cervical spondylosis, spondylolisthesis, spinal osteophytosis, vertebrobasilar insufficiency.

• Psychiatric: mental disease, mental illness, mental disorders (including: mania, schizophrenia, anxiety, depression, bipolar disorder).

• Medication: prescribed drug, prescribed medication, pharmaceutical preparations, therapeutic uses (including: anticonvulsants, neuroleptics, antipsychotics, antidepressants, psychotropics, sedatives, tranquillisers, benzodiazepines), antihistamines, insulin.

Inclusion and exclusion criteria

We limited findings to publications with an abstract in English, and excluded papers that related solely to vocational driving; those for which the health outcome was a consequence of rather than a risk factor for injury; those concerned only with alcohol or drug abuse, and those that did not conduct an analysis of accident or injury risks (or provide enough data to derive estimates of risk), including case-only series and studies solely of impaired performance.

Data abstraction and quality assessment

All of the procedures were replicated independently by KTP and ECH, and differences were resolved by consensus. Abstracts were examined, duplicates and irrelevant hits were eliminated, and paper copies then obtained of all primary research papers and reviews. We checked the reference lists of retrieved papers for supplementary relevant material.

For each primary research paper that was finally retrieved, we abstracted details of the study populations, setting, design, exposure comparisons, strategies for assessment of exposure(s) and outcome(s), response rates, confounders considered, and estimates of effect. Some papers featured numerous risk estimates for the same subcategories of injury and exposure: in these circumstances we selected the risk estimates that were the most fully adjusted for confounding. Where papers provided frequencies but not estimates of relative risk (RR) we calculated odds ratios (ORs) with exact 95% confidence intervals (95% CIs) using STATA software.

We also formed a subjective judgement on the quality of information in each paper (“quality rating”) taking into account limitations of design, potential for bias or confounding, and power to detect important associations. Studies were ranked higher if they were well powered, employed a representative sampling frame, achieved a high response rate, were prospective, controlled adequately for confounding, and had assessed exposures and outcome independently and by objective means. We rated each of these qualities individually; some of the components of our decision-making are summarised below. We also formulated an overall assessment on a four-point scale. (This did not reflect a simple sum of each individual score but a judgement informed by them.)

Confounding and effect modification

The potential for important confounding depends on the RR associated with a confounder, its prevalence and the likelihood that it might vary importantly between groups with contrasting exposures. Additionally, some factors may act as effect modifiers. Based on our understanding of risk factors for occupational accidents and occupational injury, the factors that should be allowed for in assessing confounding/effect modification are: (1) age, (2) sex, (3) location, (4) time period, (5) occupational demands/job activities, (6) job experience (years in work), (7) weekly working hours, and (8) alcohol consumption. We rated control of confounding as “excellent” (+++) if analysis and/or design allowed for seven or eight of these items; as “good” (++) if it covered five or six; as “moderate” (+) if it controlled for four of them; and as “poor” if it covered three or fewer of these items (−).

Bias

Two categories of bias need to be distinguished — “inflationary” bias (bias that could cause important overestimation of RR) and bias that could cause elevated RRs to be underestimated (bias to the null or negative bias).

Inflationary bias may arise from non-independent assessment of exposures and outcomes or from measurement error. Thus, concern arises where blinding is insufficient, or when exposure and outcome are self-reported together and in retrospect (a common design feature of the cross-sectional studies we found). Inflationary bias from measurement error is a concern when the timing of exposure relative to injury is unclear and the exposure is liable to change as a consequence of occupational injury (eg tranquilliser use, low mood), or perhaps be brought to attention through injury (eg poor vision).

Bias towards the null is of more concern where there is simple non-differential misclassification of exposure or outcome — as might arise, for example, when health limitations are assessed in vague non-objective terms. Negative bias can arise from the “healthy worker” effect and the migration of workers with health limitations to less hazardous jobs; we rated this of lower concern when analyses were stratified by or otherwise controlled for occupational activity.

We rated the potential for inflationary bias as “high”, “possible” or “low”, and that of bias leading to an underestimate of increased RRs as “possible” or “low”.

Sampling

We assessed whether the sampling frame and procedures were clearly stated, whether inclusion and exclusion criteria were explicit, and whether we could track and account for all of the subjects from the description given. Findings were graded on a three-point scale.

Exposure assessment

Some studies employed objective quantitative measures of exposure (eg measured level of hearing loss). We rated these more highly, especially where they provided exposure–response information.

Response rates

We calculated effective response rates for the analyses of interest (focusing for the cohort studies on response at follow-up), and we rated response rates of ⩾85% as “excellent” (+++), of 75–84% as “satisfactory” (++), of 50–74% as “fair” (+), and of <50% as “poor” (−).

Completeness of reporting

Incomplete reporting sometimes impaired our capacity to assess overall quality. In reaching the final rating, we assumed that missing items did not meet the criteria we proposed.

Meta-analysis

We considered the scope for meta-analysis for studies with sufficiently similar definitions of exposure (illness) and outcome (accident event), but in practice found these too limited to warrant a pooling of risk estimates.

RESULTS

Following elimination of duplicates and non-English publications we identified 760 potentially relevant abstracts. Assessment of these 760 abstracts allowed us to exclude 515 published papers that did not permit an analysis of accident risks (including 155 case-only series, 114 papers in which the health condition followed rather than preceded injury, and 70 with a sole focus on alcohol or substance misuse). We retrieved 61 papers and added to these 16 other candidate papers, identified from a perusal of reference lists. However, among the 77 papers read in full, 4 were reviews, 10 defined the injury outcome and/or health exposure inadequately, and on closer scrutiny 25 did not allow RRs of occupational injury to be derived in relation to the study exposures. Thus, finally, 38 research papers (33 independent studies^7–44) satisfied our selection criteria of which 15 papers (11 studies) were set in agricultural communities.

The main design features and our quality assessment of the final selection are presented in online supplementary table 3. Altogether, 16 independent studies (18 published papers) were of cross-sectional design, 13 were case-control studies (16 papers) and 4 were prospective cohort studies. One paper24 duplicated information from earlier studies,^{21 22} and does not appear in later tables or calculations.

Several eligible papers considered more than one health problem, and as a result we identified 19^{8 10 15 16 20–23 25 26 31 35–39 41 43 44} papers on hearing problems, 14 papers on visual problems,^{8 16 21–23 25 31 35–38 40 43 44} 15 papers related to mental health,^{11 13 14 17 21 22 28 33 35–38 40 43 44} 15 papers related to other health problems (musculoskeletal, cardiovascular, epilepsy, diabetes and allergy and asthma),^{8 9 12 15 16 20 23 29 30 35–38 40 42} and 12 papers related to medication.^{7 9 18 19 25 27 29 30–32 34 40}

Most investigations took as their outcome accidental injury in the past 12 months, although outcome definitions were heterogeneous, varying for example in the extent of injury, the residual limitation and the involvement or otherwise of medical aid; in four studies^{10–12 18} the outcome was defined as an accident rather than as an accidental injury or accident requiring medical attention.

Several studies employed an independent assessment of exposure^{10 13 25 39 42} or outcome^{18 28 30 31} or both,^{12 29 40} but for the most part, exposures and outcome were ascertained by self-report at a single time point. Few of the cross-sectional and retrospective studies established by design that exposure preceded outcome, and we rated the potential for inflationary bias as “high” for at least some comparisons in 14 of the studies.^{7 11 14 15 17 19–22 25 28 33 35–38} Confounding was addressed in various ways (restriction, matching, stratification, regression modelling), but 15 of the studies^{9 13 15 16 18 21–24 27 29 30 32–34 42 44} failed to control for five or more of the eight factors suggested by us as relevant.

We rated the overall quality of information as excellent (++++) in two studies,^{29 40} as useful but with some important limitations in seven studies (+++),^{10 23 25 31 35–37 43 44} as moderately informative (++ or +(+)) in 10 studies^{8 11 16 20–22 26 28 30 39 41} and as limited in 14 studies (+).^{7 9 12–15 17–19 27 32–34 42}

The main findings are presented in tables 1 and 2 and online supplementary tables 3–5. Each table is organised first by exposure category and study design and then alphabetically by first author.

Impairments of hearing

The relation between risks of accidental injury and hearing problems was considered in 15 studies, including three of cohort design (table 1). Exposure assessment was mostly based on self-report, although three studies made use of the judgement of an assessor^{10 23 25} and in three studies, pure tone audiometry was measured.^{31 39 41}

None of the various studies on hearing impairment were classed as having a high potential for inflationary bias. In most comparisons, moderately positive associations were reported (OR ⩾1.5) with RRs sometimes exceeding 2.0^{10 16 23 31 35–39}; and most of the studies of higher quality were compatible with a rough doubling of risks.^{23 31 35–38} In the largest one, based on over 76 000 subjects from the cross-sectional National Health Interview Survey,23 odds were raised over 2-fold among those considered deaf by the interviewer and 1.6-fold in those with self-reported hearing impairment; a measured hearing loss of 25 decibels hearing loss (dBHL) in the better ear was associated with an OR of 1.6 in a cohort study,41 and a 20 dBHL was associated with an OR of 1.9 in one case-control study.31 In another case-control study the odds of accidental injury were more than doubled for a binaural hearing loss of 20%–54%, although findings were not significant at the 5% level.39 There were some indications in this last investigation that hearing loss in a noisy work environment raised risks even further. In a cross-sectional study of physician-assessed hearing disorder, significantly increased ORs were found in relation to injuries requiring hospitalisation or prolonged sick leave, although there was little evidence that risk varied by type of incident causing injury10; the nature of the hearing disorder and the criteria employed in diagnosis were unstated.

In only two of the studies was there an attempt to distinguish risks by sub-category of external cause^{10 35–37} and none reported outcomes by anatomical site or nature of injury.

On balance we assess the evidence as favouring a moderately higher risk of accidental injury in those with hearing impairment; a few studies included objective measurement of hearing loss, and these tend to support this interpretation.

Impairments of vision

We identified 10 studies on problems of vision (table 1), including two studies of cohort design.^{43 44} Three of the papers made use of an assessor’s judgement (“blind”, physician-determined poorly corrected vision disorder, and medically diagnosed eye disorder),^{23 25 40} but most took as their exposure definition self-report of visual difficulty or wearing glasses.

There were few positive findings of note, although Zwerling et al reported associations with poor self-rated vision in one cross-sectional survey (OR 1.5 in non-farmers, 3.1 in farmers),^{21 22} and with an interviewer’s opinion that the participant was blind in another (OR 3.2).23 In one cohort study, also by Zwerling et al,44 the OR for accidental injury was 1.45 (p>0.05) in those with self-reported poor sight; in a much smaller study by the same group,43 no injuries were reported in the group with a “vision problem”. A case-control study that took medically diagnosed eye disorder as its risk factor reported an OR of 1.2 (p>0.05) but provided no breakdown by diagnostic subcategory.40 In several analyses, ORs were less than 1.0 (although none were significantly so at the 5% level).

In summary, we found little evidence that impairments of vision increase risks of occupational injury, but also few investigations with a focus on well-defined eye pathology.

Poor mental health

Findings in relation to mental health were considered in 11 studies and results were mixed (table 2). In many studies lower confidence limits exceeded one and several studies indicated ORs ⩾1.5, or even higher in certain subgroups. However, we classed more studies as prone to inflationary bias in this than for other categories of health problem.^{11 14 17 21 22 28 33 35–38}

Approaches to exposure definition varied. The CES-D scale for depression was a popular instrument,^{14 17 21 22 35–38 43 44} although with differing cut-off points chosen to define high exposure; in addition, three other screening instruments for minor psychiatric disorder and emotional instability were employed,^{11 28 33} as well as a physician’s diagnosis of depression35^–38 or neurotic disorder.40 One small study of limited quality was found on schizophrenia.13

Among studies that used the CES-D depression scale, the three largest suggested only modest RRs (OR <1.5 for accidental injuries overall, all with p<0.05).^{17 21 44} In the three studies of highest quality, ORs ranged from 1.37 to 3.22,^{35–38 43 44} the extremes representing the two investigations which had a prospective design. Self-report of doctor-diagnosed depression carried an OR of 1.82 (p<0.05) for all injuries in a case-control study of farmers by Sprince et al,36 and an OR of 2.37 (p<0.05) for fall-related injuries in the same study group38; but an OR of 1.07 was found in a second case-control study that linked records of hospital attendance for injury and prescribed medication for international classification of diseases (ICD)-defined neurotic disorder.40

On balance, we assessed the evidence as favouring a higher risk of injury in those with emotional problems, while not firmly establishing this to be so.

Other long-term health conditions

Online table 4 summarises our findings in relation to five other categories of long-term health problem. Musculoskeletal symptoms were assessed in six studies, largely based on self-report of regional pain or “arthritis”.^{8 15 20 24 35–38 40} ORs were generally ⩽1.5, and in a single high-quality paper based on physician’s diagnosis of osteoarthritis the estimated OR was close to unity.40 However, in one study it was significantly raised for self-report of joint discomfort (OR 2.56),15 while Sprince et al found risks of occupational injury in those with self-reported arthritis or rheumatism to be elevated 2- to 3-fold in sub-analyses related to specific types of injury (falls and injury with livestock).37^–38

For cardiovascular disease (self-report of heart disease,^{20 35–38} self-report of high blood pressure,20 or doctor diagnosis of these disorders40) the evidence base was sparse but did not point to elevated risks.

We found four studies on epilepsy.^{12 16 23 42} ORs were raised 1.5- to 2.5-fold, although findings were not significant at the 5% level in three of the four studies, including a very large cross-sectional study by Zwerling et al.23 Two of the remaining studies, rated of lower quality, failed to account for a matched design in their analysis.^{12 42}

We identified three studies concerning diabetes or prescribed diabetic medication.^{23 29 40} The largest study found a moderate elevation of risk (OR 1.47),23 as did a well-conducted case-control study linking recent hypoglycaemic prescription with medically recorded injuries (OR 1.3–1.4).29 Risk of accidental injury was also somewhat higher for physician diagnosis of diabetes in another record linkage study by Voaklander et al.40

Finally, we identified four studies on allergy, hay fever and asthma.^{9 20 30 35–38} In a cross-sectional study by Bunn et al9 there was a trend of increasing injury risk with increasing severity of doctor-diagnosed allergy, although numbers in the analysis were unclear. In a small study of self-reported allergy there was no such increase in risk of injury.20 In a large case-control study, self-reported doctor-diagnosed asthma showed a moderate association with injury risk overall (OR 1.6),36 with a higher risk in a sub-analysis related to injuries from livestock (OR 2.46, p<0.05)37, and in a sub-analysis confined to fall-related injuries (OR 2.27)38; while in a second large case-control study, there was little association between acute traumatic injury and physician-diagnosed nasal allergy.30

The papers on epilepsy and diabetes, although consistent with a small increase in risk of accidental injury, provide a limited evidence base on which to draw conclusions. Those on allergy were few and inconsistent.

Medication

We found several papers on medication — related to use of anxiolytics, hypnotics and sedatives,^{18 19 25 29 31 32 34 40} antidepressants,^{19 29 34} antipsychotics,29 or otherwise psychoactive medicines,^{6 18} and other drugs with sedative potential (narcotics^{29 40} and antihistamines^{9 27 29 30}).

Voaklander et al40 found that prescription of anxiolytics, sedatives or hypnotics in the preceding 30 days was associated with a 3-fold increase in odds of hospital attendance with work-related injury, whereas in a study of similar design by Gilmore et al29 ORs were much lower (0.8 in men and 1.5 in women). Two other studies favoured a more than doubling of risk,^{19 25} although both had the potential for inflationary bias through reverse causation — in Wadsworth et al,19 for example, the taking of sleeping pills related to the 14 days prior to questioning, whereas injuries might have occurred up to a year beforehand.

The study by Gilmore et al29 found little evidence of elevated risks in those taking antidepressants, antipsychotics or narcotics; and two other studies of lower quality found limited (OR 1.5)7 or no effect18 from psychoactive medication in general.

Hanrahan et al30 reported that antihistamines in the prior 2 weeks raised the odds of accidental injury almost 3-fold in an adjusted analysis that included a term for interaction between use of antihistamine and age. However, there were no important differences between injury cases and referents in the crude prevalence of sedative antihistamine use (9% vs 8%, respectively). Bunn et al9 found a cross-sectional relation between self-report of injury and sedative antihistamine use in the past 12 months (prevalence ratio 1.6, p<0.05); but the prevalence of injury was similar in those using non-sedative antihistamines, pointing if anything to an effect from hay fever rather than its treatment.

Finally, Voaklander et al40 found moderately positive associations with prior use of non-steroidal anti-inflammatory drugs.

In summary, most of the data we found on medication and injury risk related to drugs with sedative potential. Findings were compatible with a moderate increase in risks, although not wholly consistent.

DISCUSSION

Our review suggests that some chronic health conditions and their treatments, including impaired hearing, neurotic illness, diabetes, epilepsy and use of sedating medication may raise the risks of occupational injury to a moderate degree, the evidence base being most complete in relation to hearing (15 studies). However, the most notable finding is an apparent shortage of good-quality evidence. Thus, for example, studies of hearing impairment seldom employed objective measures of hearing loss; those on vision did not present risks by specific categories of eye disease and did not employ a quantitative measure of impairment; we found no evidence on major categories of psychiatric illness such as bipolar disorder and mania; for some common important health outcomes including diabetes, epilepsy and cardiovascular disease, the evidence base was remarkably thin; and first injury was seldom distinguished from recurrent injury risk. Moreover, few studies attempted to distinguish risks by category of external cause (eg fall, injury from machinery) or by anatomical site or nature of injury (eg a fractured femur, burn to the hand). The studies that we did identify tended to have important limitations, including potential for confounding and inflationary bias, and a frequent lack of clarity regarding the timing of illness relative to injury; there were few prospective investigations. Finally, health-related selection into and out of jobs may have led to residual confounding by work activity, insofar as studies tended to control for this factor only crudely, at the level of occupational title. Apparently protective effects of some health problems in some studies may have arisen from such selection. Thus, for example, the finding of an association with impaired hearing but not impaired vision could reflect an earlier withdrawal from certain hazardous work in those with overt problems of seeing than in those with insidious loss of hearing.

Our search had limitations too and may not have been fully comprehensive. We did not assess the grey literature or consult experts in accident research or review the research abstracts of conferences. However, the search encompassed the three major biomedical bibliographic databases; we were thorough in the search terms we employed; and we checked other reviews and their bibliographies for relevant material. It seems unlikely, therefore, that a major volume of high-quality research has been overlooked.

The relative shortage of information on occupational risks can be contrasted with a more extensive literature on health impairments and RTAs (including some studies of vocational drivers). In a previous review of this topic45 we identified several studies employing specific measurements of visual performance (visual acuity, field of vision, binocularity and contrast sensitivity), and covering several specific eye diseases (cataracts, glaucoma and macular degeneration); reduced field of vision was consistently associated with risk of accidental injury, with estimated RRs ranging from 1.9 to 22.0 for a >40% reduction.3^–5 We also identified some 14 primary research studies on epilepsy and RTAs, including seven papers of cohort design; in general these suggested that the excess risk, if any, was modest, and that patients who experienced reliable auras or complied with anti-epileptic treatment or had their last seizure some time ago did not have an increased risk of RTA^{46 47}; in one cohort study an increased risk of RTAs existed for patients with generalised and complex seizures but not for patients with simple seizures.48 The present review failed to identify any papers relating to occupational injuries that covered common health problems in comparable depth.

The information gap is both surprising and urgent to fill. There is a pressing need, particularly in the context of an ageing workforce, for more and better targeted research to ensure that health-related decisions on job placement are evidence-based. Future research should define exposures and outcomes in greater detail, while ensuring by design that the former precede the latter.