Results of the search

The search for this update yielded 4021 records after removal of duplicates. From these, we selected 63 articles for full‐text assessment, plus four identified through other sources. Of these 67 articles, we excluded 56. The remaining 11 articles described four trials; one was new (DanMONICA 1982), one was already included but had new results (Inter99 1999), and two were ongoing trials. We did not identify any additional trials by searching reference lists and through citation tracking. In the previous version of this review, we included 16 trials (Ebeltoft 1992; Family Heart 1990; Göteborg 1963; Göteborg 1970; Inter99 1999; Kaiser Permanente 1965; Malmö 1969; Mankato 1982; New York 1971; Northumberland 1969; OXCHECK 1989; Salt Lake City 1972; South‐East London 1967; Stockholm 1969; Titograd 1971; WHO 1971), but two trials did not report results (New York 1971; Titograd 1971). With one new trial identified (DanMONICA 1982), we therefore had 17 trials in all, and 15 trials with data for our updated analysis (Figure 1).

Figure 1

---

---

Study flow diagram

Included studies

We included 17 trials, 15 of which reported outcome data.

The 15 trials with data varied in size from 533 randomised participants in Northumberland 1969, to 61,301 in Inter99 1999. The total number of participants was 251,891 with 87,412 allocated to health checks and 164,479 to a control group. Eleven trials with 233,298 participants reported a total of 21,535 deaths (DanMONICA 1982; Ebeltoft 1992; Göteborg 1963; Göteborg 1970; Inter99 1999; Kaiser Permanente 1965; Malmö 1969; OXCHECK 1989; South‐East London 1967; Stockholm 1969; WHO 1971). The length of follow‐up for total mortality varied from 4 to 30 years, and it also varied for other outcomes. The trials that did not report mortality were often small (Mankato 1982; Northumberland 1969; Salt Lake City 1972), with the exception of the British Family Heart study (Family Heart 1990), which included 12,924 participants.

The setting was general practice in five trials (Family Heart 1990; Ebeltoft 1992; Northumberland 1969; OXCHECK 1989; South‐East London 1967), medical centre/research centre in nine trials (DanMONICA 1982; Göteborg 1963; Göteborg 1970; Inter99 1999; Kaiser Permanente 1965; Malmö 1969; Mankato 1982; Salt Lake City 1972; Stockholm 1969), and the workplace in one trial (WHO 1971). As per our inclusion criteria, they included people that were not selected for diseases or risk factors. Four trials randomised households or couples (Family Heart 1990; OXCHECK 1989; Salt Lake City 1972; South‐East London 1967), one randomised factories (WHO 1971), and ten randomised participants. Three trials were conducted in the USA (Kaiser Permanente 1965; Mankato 1982; Salt Lake City 1972), and twelve were conducted in Europe (DanMONICA 1982; Ebeltoft 1992; Family Heart 1990; Göteborg 1963; Göteborg 1970; Inter99 1999; Malmö 1969; Northumberland 1969; OXCHECK 1989; South‐East London 1967; Stockholm 1969; WHO 1971).

The interventions offered can be broadly classified into two categories: screening focused on cardiovascular risk factors with a strong lifestyle intervention component, and broad screenings using many tests (often called multiphasic screening in older publications) but often without an important lifestyle intervention component. The very broad type of screening was mainly seen in trials that started in the 1960s and 1970s. Five trials included screening for cancer. The tests used were chest radiographs (Göteborg 1963; Malmö 1969); chest radiographs and faecal occult blood testing (South‐East London 1967); chest radiographs, mammography and cervical smears (Salt Lake City 1972); and chest radiographs, sigmoidoscopy, mammography and pelvic examinations (Kaiser Permanente 1965). One trial used abdominal ultrasound (DanMONICA 1982). See Table 1 for an overview of the interventions used. Six trials included a physical examination by a physician (Göteborg 1963; Kaiser Permanente 1965; Malmö 1969; Northumberland 1969; South‐East London 1967; Stockholm 1969), while nine trials did not (DanMONICA 1982; Ebeltoft 1992; Family Heart 1990; Göteborg 1970; Inter99 1999; Mankato 1982; OXCHECK 1989; Salt Lake City 1972; WHO 1971).

The uptake in the first screening round ranged between 50% (Mankato 1982) and 90% (Ebeltoft 1992) with a median of 80%. Kaiser Permanente 1965 did not use screening rounds at specific intervals but urged the intervention group repeatedly by written invitations and phone calls to utilise a pre‐paid health check.

We chose to label the studies with the year of trial start, instead of year of publication, for the following reasons: 1) year of publication would make the data look much younger than it is, especially in trials with very long follow‐up, 2) results were often scattered in several papers in different years, whereas year of trial start is more well‐defined. Other citations are labelled in the standard fashion with year of publication.

Excluded studies

We excluded 56 studies found in the updated search (Characteristics of excluded studies). The list of excluded studies (Characteristics of excluded studies) also includes one important but not eligible trial not found in the search (Lindholt 2017). In the previous version of the review, two articles were awaiting assessment (Brett 2012; Stickler 2000). These have been excluded.

Allocation

Seven trials used a genuinely random method for generating the randomisation sequence (DanMONICA 1982; Ebeltoft 1992; Göteborg 1970; Inter99 1999; Mankato 1982; OXCHECK 1989; Stockholm 1969). We could not determine how six trials generated the sequence (Family Heart 1990; New York 1971; Salt Lake City 1972; South‐East London 1967; Titograd 1971; WHO 1971). Four trials used allocation methods such as date of birth (Göteborg 1963; Kaiser Permanente 1965; Malmö 1969; Northumberland 1969), but these trials included participants through lists or registers and allocated them all at once before making any contact with them, and we therefore judged the risk of selection bias to be low.

We judged allocation to be adequately concealed in 14 trials (DanMONICA 1982; Ebeltoft 1992; Family Heart 1990; Göteborg 1963; Göteborg 1970; Inter99 1999; Kaiser Permanente 1965; Malmö 1969; Mankato 1982; Northumberland 1969; OXCHECK 1989; South‐East London 1967; Stockholm 1969; WHO 1971). It was unclear in three trials (New York 1971; Salt Lake City 1972; Titograd 1971).

We thus judged 11 trials as likely to be free from selection bias (DanMONICA 1982; Ebeltoft 1992; Göteborg 1963; Göteborg 1970; Inter99 1999; Kaiser Permanente 1965; Malmö 1969; Mankato 1982; Northumberland 1969; OXCHECK 1989; Stockholm 1969). We could not rule out selection bias in six trials. Five trials did not describe sequence generation (Family Heart 1990; New York 1971; Salt Lake City 1972; Titograd 1971; WHO 1971), and South‐East London 1967 used a matching procedure, which was unclearly described, for randomisation, and the sizes of the groups varied between publications.

Blinding

Incomplete outcome data

Selective reporting

We found nine trials to be at low risk of reporting bias (DanMONICA 1982; Family Heart 1990; Göteborg 1963; Göteborg 1970; Inter99 1999; Malmö 1969; Mankato 1982; OXCHECK 1989; WHO 1971), in four trials the risk was unclear (Ebeltoft 1992;; Northumberland 1969; Salt Lake City 1972; Stockholm 1969) and in four trials the risk of reporting bias was high: Kaiser Permanente 1965 collected data on surgery, prescriptions, and reasons for hospitalisation but did not publish them. They also collected and reported results on new diagnoses in early publications but not for the planned study period; South‐East London 1967 collected but did not report data on referrals, prescriptions, and investigations carried out; and New York 1971 and Titograd 1971 have never published any results. A funnel plot for mortality did not suggest publication bias (Figure 4).

Figure 4

---

---

Funnel plot of comparison 1. Health checks versus control, outcome: 1.1 Total mortality.

Other potential sources of bias

Four trials had a design that could favour the screening group (Göteborg 1963; Göteborg 1970; Malmö 1969; WHO 1971). In these trials, conditions identified at screening were treated and followed at a special clinic or by the researchers whereas participants in the control group used their normal physicians. However, this could also increase possible iatrogenic harm.

Screening of the control group (contamination) would dilute both the beneficial and the harmful effects of the intervention. Only two trials assessed the number of participants in the control group having health checks. In Kaiser Permanente 1965, after 16 years, the mean number of health checks in the control group was 2.8 compared with 6.8 in the screening group, and 36% of the control group had not had a health check compared to 16% of the screening group. However this result cannot be generalised to the other trials or other populations, mainly because the participants were all members of the same health plan with access to the same high‐profiled multiphasic health screening. Also, screening has long been more popular in the USA than in Europe. In the South‐East London Screening Study (South‐East London 1967), there was very little interest in screening among the participants in the control group, and none were screened for the first five years (Trevelyan 1973). However, the control group was offered screening after five years, which biased the nine‐year results towards no effect.

The British Family Heart Study (Family Heart 1990), used both an internal and an external control group in order to investigate contamination. They found similar results when comparing with either control group indicating that contamination was not a big problem. In Ebeltoft 1992, which was set in a small town, the trial authors noted that the trial appeared to have a large positive influence on the health behaviours of the control group (Lauritzen 2012). Also, the control group was offered screening after five years while some data were collected for eight years. Mankato 1982 was conducted during a health promotion campaign, which may have diminished the effect of the intervention.

In summary, we found seven trials with a low risk of contamination (DanMONICA 1982; Göteborg 1963; Göteborg 1970; Inter99 1999; Malmö 1969; Stockholm 1969; WHO 1971), six trials in which it was unclear (Family Heart 1990; New York 1971; Northumberland 1969; OXCHECK 1989; Salt Lake City 1972; Titograd 1971), and four trials with a high risk of contamination (Ebeltoft 1992; Kaiser Permanente 1965; Mankato 1982; South‐East London 1967).

Two trials randomised people who had returned an initial questionnaire on health and lifestyle (Ebeltoft 1992; OXCHECK 1989). This limited the external validity because of self‐selection of people with an interest in health and lifestyle (Pill 1988; Waller 1990).

Total mortality

Eleven trials reported total mortality (DanMONICA 1982; Ebeltoft 1992; Göteborg 1963; Göteborg 1970; Inter99 1999; Kaiser Permanente 1965; Malmö 1969; OXCHECK 1989; South‐East London 1967; Stockholm 1969; WHO 1971).The median length of follow‐up was 10 years and the range was 4 to 30 years.The median event rate in the intervention and control groups combined was 7% and the range was 2% (OXCHECK 1989), to 36% (DanMONICA 1982), reflecting the different lengths of follow‐up (Table 2).

Health checks have little or no effect on total mortality (RR 1.00, 95% CI 0.97 to 1.03; 11 trials; 233,298 participants and 21,535 deaths; high‐certainty evidence, I² = 0%; Analysis 1.1). Subgroup and sensitivity analyses did not alter the results.

Disease‐specific mortality

Morbidity

New diagnoses

In addition to conditions identified through the screening itself, screening might increase diagnostic activity between scheduled screenings due to increased physician contact in relation to follow‐up visits or due to a lowered threshold for consulting a physician. Cumulative rates of new diagnoses over time in the screened and unscreened groups would allow an assessment of the full effect of screening on diagnostic activity. However, only one trial reported such results (Kaiser Permanente 1965), and only for the first six years. In a 40% sample, Kaiser Permanente 1965 found a sharp divergence in the mean annual number of new diagnoses per participant immediately after the intervention started, with the differences being statistically significant each year. By adding the results for each year we found a mean number of new diagnoses per participant of 4.3 in the screening group and 3.6 in the control group. This corresponded to a 20% increase. The trial lasted for 16 years but follow‐up for new diagnoses was not continued.

Four trials reported their findings at the first screening of the intervention group but without comparisons with the control group over time. South‐East London 1967 found an average of 2.3 diseases per participant at the first screening. Of these, 53% were not previously known. Ebeltoft 1992 reported the percentage of participants with abnormal findings prompting health advice at the initial screening to be 76%. The most common reasons were raised CO concentration in expiratory air in smokers (37%), low physical endurance (30%), poor hearing (19%), poor sight (12%), and being overweight (16%). Increased cardiovascular risk was found in 11%, hypercholesterolaemia in 10%, hypertension in 10%, and elevated liver enzymes in 13%. Salt Lake City 1972 found a total of 2031 abnormalities in 384 people screened. This trial used very broad biochemical screening.

In summary, health checks may increase the number of new diagnoses (low‐certainty evidence).

Admission to hospital

Five trials reported hospitalisation using different measures, for example, admission rates, number of people admitted once or more, or number of days in hospital.

Kaiser Permanente 1965 reported the mean number of days in hospital over 18 years of follow‐up. The results were 10.0 days in the intervention group and 10.4 days in the control group. Roughly one third of participants had missing data. South‐East London 1967 reported the number of participants admitted to hospital once or more during nine years of follow‐up, RR 1.04 (95% CI 0.96 to 1.13). The amount of missing data was unclear but was probably low. Malmö 1969 also studied the number admitted once or more and found similar results, RR 1.05 (95% CI 0.92 to 1.20). There were 3% to 5% missing data. Salt Lake City 1972 compared hospitalisation rates before and after the intervention and did not find an effect, but they did find an effect on the number of nights in hospital in one of three subgroups, which was an unreliable result due to biased exclusions after randomisation. Ebeltoft 1992 compared admission rates in the two intervention groups with the control group and did not find an effect after eight years, rate ratio of 0.91 (95% CI 0.63 to 1.32). They also compared the random sample invited to participate in the trial with all not invited and found similar results, rate ratio of 0.97 (95% CI 0.80 to 1.18). There were 5% missing data.

In summary, health checks may make little or no difference to admission rates, number of people admitted once or more, or number of days in hospital (low‐certainty evidence).

Disability

Three trials investigated the effect on disability. Kaiser Permanente 1965 found that after 16 years 31% of the screening group and 30% of the control group reported total or partial disability on a questionnaire. Attrition was roughly one third and response rates around 75%, which left only half of the people randomised in this analysis. South‐East London 1967 found that 2.5% in the screening group and 1.8% in the control group reported major disability after five years. There were between 40% and 50% missing data in this analysis. Salt Lake City 1972 compared the number of disability days before and after the intervention and did not find an effect.

In summary, health checks may make little or no difference to disability (low‐certainty evidence).

Worry

Only two trials reported relevant results, using scales measuring psychological distress.

Ebeltoft 1992 used the General Health Questionnaire (GHQ‐12) at baseline and after one and five years. A decrease in score indicates a beneficial effect of the intervention. After one year, the change from baseline in the screening groups was an increase of 0.05 and in the control group a decrease of 0.16, P = 0.6. After five years, the screening group had a decrease of 0.23 and the control group had a decrease of 0.39, P = 0.73. They also investigated subgroups of smokers, overweight participants, people who were informed of an elevated risk and people informed of no elevated risk, and did not find effects. Participation was 79% after five years.

South‐East London 1967 used the Middlesex Hospital Questionnaire on a subset of participants after five years. In the anxiety domain of the scale, the trial authors found lower scores in the intervention group among men (lower scores are better). When pooling men and women, we found a mean score of 4.14 (standard deviation (SD) = 3.38, 602 participants) in the intervention group and 4.48 (SD = 3.63, 572 participants) in the control group, P = 0.10 (t‐test, equal variances). In the other domains assessed with this scale ('phobic', 'obsessional', 'somatic', 'depression', 'hysteria'), there were no effects. Follow‐up was roughly 90%.

In summary, health checks may make little or no difference to worry (low‐certainty evidence).

Self‐reported health

Four trials reported self‐reported health.

South‐East London 1967 found that after five years 53.6% of the screening group and 56.5% of the control group reported good or excellent health in the preceding two weeks (Chi² = 3.274, P = 0.07).

Ebeltoft 1992 used a five‐point scale at baseline and after five years. After five years, 70% and 72% of the two intervention groups reported good or excellent health compared to 71% of the control group. Data on change from baseline were only available in a graph. This showed that approximately 12% in the intervention groups had an improvement in self‐reported health compared to approximately 20% in the control group.

In Family Heart 1990, 79.5% of the screening group and 75.7% of the internal control group reported good or excellent health after one year. This analysis used last observation carried forward for missing data. The pooled difference, taking into account the 13 different practices, was 3.8% in favour of screening, P = 0.004.

Inter99 1999 used SF‐12 and found slower deterioration of both physical and mental health components in the intervention group. For mental health, the difference after five years was approximately 2 on a 100‐point scale, where 50 is the mean of a reference population with a SD of 10. The effect was even smaller for physical health but was difficult to assess because of baseline imbalances in scores. The trial authors found indications of biased non‐response.

In summary, health checks may slightly improve self‐reported health (low‐certainty evidence).

Referrals to specialists

Only one trial (Ebeltoft 1992), reported on this outcome, but we could not use the results in our analysis. The trial authors only had data from 1995 to 1999 but the screening took place in 1992 to 1993 (intervention groups screened), and 1997 (intervention groups and control group screened). Thus the expected increase in referrals following the intervention was not included in the analysis and any contrast between groups would be diluted by the 1997 screening. The trial authors made two comparisons and did not find effects in either analysis. When comparing the screening and control groups, the rate ratio was 1.04 (95% CI 0.85 to 1.26). When comparing the random sample invited to participate in the trial versus all eligible people not invited, the rate ratio was 0.94 (95% CI 0.84 to 1.06).

In summary, it is uncertain whether health checks increase or reduce referrals to specialists, as the certainty of this evidence is very low.

Non‐scheduled visits to general practitioners

Five trials reported physician visits. The length of follow‐up was between one and nine years, with missing outcome data ranging between 5% (Ebeltoft 1992) and 51% (Salt Lake City 1972).

Kaiser Permanente 1965 found a mean number of physician visits of 16.0 in both groups after five years, not including the screenings themselves. The results were reported without measures of uncertainty and data on this outcome were collected from a 20% subsample.

South‐East London 1967 did not find an effect on the mean annual number of physician visits. It was not clear whether the screening visits were included in this, and we cannot tell whether the results were from the five‐year or nine‐year follow‐up. We excluded participants who left the study before one year from the analyses (14% from the screening group and 13% from the control group).

Northumberland 1969 found an average number of consultations per participant of 5.4 in the screening group and 5.0 in the control group over 1½ years. This did not include the screenings themselves. When adding the screenings the results were 6.3 in the screening group and 5.0 in the control group. The trial authors did not specify the type of health check, and there was a high risk of detection bias, as the allocation was noted on the front cover of the participant's record.

Salt Lake City 1972 did not find effects after one year, but this result was unreliable due to biased exclusions. The screening visits were not included in the analysis.

Ebeltoft 1992 found an increased rate of physician visits after five years in the screening plus health discussion group compared to the control group, rate ratio of 1.15 (95% CI 1.02 to 1.31), but not in the screening only group compared to controls, rate ratio of 1.01 (95% CI 0.89 to 1.15). When comparing all those invited to participate in the trial with all not invited, the rate ratio was 1.01 (95% CI 0.93 to 1.10). However, this comparison included screening of the control group in 1997, diluting any differences between groups. The trial authors found a downward trend in the rate ratio over time favouring the intervention, but in the absence of an overall effect this is not a relevant observation. It likely reflects the initial increase in visits generated by the screenings themselves, which gave a high starting point for the trend analysis.

In summary, health checks may make little or no difference to the number of physician visits (low‐certainty evidence).

Additional diagnostic procedures required due to positive screening tests

We did not find any trials that reported this outcome.

Kaiser Permanente 1965 reported the mean number of laboratory tests per participant after five and 10 years, based on a 20% sample. After five years it was 23.8 in the screening group and 23.3 in the control group. The data after 10 years were not reported but the trial authors stated that there was no difference. The number of laboratory tests did not include the tests used at screening.

Prescriptions and surgery

None of the trials reported the total number of prescriptions, new drugs prescribed, or the number of operations performed. This is unfortunate since these are important factors for balancing the benefits and harms of health checks and for estimating the costs.

Five trials provided some results of relevance.

Göteborg 1970 examined random samples of the intervention group and control group and found that after 10 years of follow‐up 26.0% of the intervention group used antihypertensive medications compared to 19.6% in the control group (Chi² = 16.41, P < 0.0001, our calculation). Kaiser Permanente 1965 reported narratively that prescription rates gathered from pharmacies showed a non‐significant trend towards increased prescription in the screening group, but only analysed data from years six and seven. Ebeltoft 1992 presented data on self‐reported use of selected types of drugs after five years. In the screening groups, 4.8% reported using blood pressure medication compared to 6.8% in the control group (Chi² = 1.42, P = 0.23, our calculation). For diuretics, the figures were 3.7% (screening), and 3.9% (control group), and for heart medication they were 0.9% (screening), and 1.0% (control). Family Heart 1990 reported narratively that there was no difference between the intervention and control groups regarding use of drugs to lower blood pressure or cholesterol, or for diabetes. Mankato 1982 reported that the proportion of participants on blood pressure medication after one year was 13.8% in the intervention group and 9.8% in the control group (P < 0.05).

In summary, it is uncertain whether health checks increase or reduce prescriptions and surgery, as the certainty of this evidence is very low.

Absence from work

Two trials reported absence from work (Kaiser Permanente 1965; South‐East London 1967). Neither trial found an effect, and neither trial reported the exact results but only mentioned their findings in a narrative.

It is uncertain whether health checks increase or reduce absence from work as the certainty of this evidence is very low.

Subgroup and sensitivity analyses

We planned and performed several subgroup and sensitivity analyses. Some of the resulting subgroups were based on very few trials but are presented for completeness (see graphs). They should be interpreted with caution and we found no convincing patterns.

For outcomes not included in the meta‐analyses we were not able to discern any patterns except that the more recent trials often had a strong focus on lifestyle interventions, often had changes in risk factors as their primary outcomes, and were designed accordingly, with shorter follow‐up (Ebeltoft 1992; Family Heart 1990; Mankato 1982; OXCHECK 1989).