Summary of findings

To our knowledge, this is the first multisite study to present data and evaluate the impact of the COVID-19 ‘lockdown’ policy in the UK on mortality and MH service activity provision. In summary, the initiation of lockdown was associated with increases in the death rate, decreases in new accepted referrals and decreases in inpatient admissions and daily caseload. In community services, sizeable increases were observed in non-f2f contacts, with decreases in f2f contacts. Caseloads decreased in CAMHS, HTT, liaison and OA services, but increased slightly in AMH and EIP services. The lift-of-lockdown transition was associated with decreased deaths, increased referrals and inpatient admissions and decreased inpatient discharges. In community settings, most services saw increases in non-f2f contacts and total contacts. There was moderate-to-high heterogeneity of these estimates across sites; only site-wide new referrals/discharges and inpatient admissions/discharges were consistent.

Impact of the COVID-19 first wave

The potential impact of the COVID-19 pandemic has been widely discussed in a general sense, focusing on the initial priorities of infection control, treatment options for severe complications and the preparedness of critical care services.25 For mental healthcare, there is a need to understand the population-level impact of both the viral infection and the social distancing being imposed by many national governments,9 26 27 as well as concerns about higher levels of population distress coupled with reduced use/availability of psychological support.28 For people with pre-existing mental disorders, there is also a concern that vulnerability to COVID-19 infection may be higher than expected, because of infection susceptibility due to physical comorbidities and because of barriers to health service access.29 30 A higher risk of relapse of mental disorders is a concern due to the stress of the pandemic itself, the stress of consequent quarantine,31 reduced access to routine outpatient visits for evaluations and prescriptions and possibly also an avoidance of health services because of the perceived risk of infection or a wish to ‘protect the NHS’. Finally, a higher risk of suicide might conceivably result from rapid social, economic and health changes.32 33 On the other hand, it is possible that the pandemic may have had positive effects on MH and the reduced use of services might reflect reduced need—for example, arising from increased family contact at home during ‘lockdown’, and/or reduced exposure to social pressures outside the home, and/or a sense of a more supportive community or shared adversity. Either way, the scale of mental healthcare changes has not yet been fully quantified beyond single-site findings, although recommendations made in China for tighter admission criteria and reduced hospital outpatient visits, among others,34 may reflect similar changes in service provision to those we report.

While it was not our intention here to investigate factors underlying the observed service changes, many are likely to be unsurprising. Clearly, a reduction in f2f clinical contacts was a likely outcome of social distancing, plus rising concern about the infection risks to patients and staff of such contacts. These were balanced by an increase in non-f2f contacts, and the only reductions after 23 March 2020 in total contacts were seen in liaison and HTT services. These also saw strongest reductions in daily caseload, which for liaison services may reflect high numbers of discharges from acute hospitals to prepare for a surge in pandemic-related demand, and possibly patient reluctance to attend emergency care. While HTT caseload reductions might reflect reduced demand, the levels and consequences of unmet need during the lockdown require further evaluation. Of interest, rates of cancelled or non-attended appointments did not change significantly for most services. On the face of it, this is reassuring, given the sizeable switch towards non-f2f contacts. However, the findings should be viewed with some caution, as it is not yet clear whether unsuccessful attempts at non-f2f contacts are recorded in the same way as a non-attended f2f appointment. Furthermore, longer-term trends beyond the rather unusual context of the ‘first-wave’ lockdown need additional evaluation.

Between-site variation

An important feature of the findings was their heterogeneity between sites. Mainly this reflected magnitude rather than direction of effect: for example, all sites saw increases in deaths around lockdown initiation (figure 1, online supplemental table 1), although IRRs varied in size. This heterogeneity may reflect site differences in patient demographics and pandemic timing across the country. IRRs for inpatient caseload changes at the same transition point may reflect differences in facilities—for example, strongest caseload reductions were seen at site A which had highest pre lockdown levels (table 1). On the other hand, reductions in HTT caseloads were weakest at site H where pre lockdown caseloads were highest and liaison caseload reductions differed substantially between sites (B and H) with similar pre lockdown caseloads; this might reflect local differences in acute hospital practice/demand and indicates scope for further investigation.

Causality considerations

It is challenging to disentangle the effect of the ongoing COVID-19 pandemic itself from those of social distancing/lockdown policies (or other psychosocial consequences) on MH service activity and mortality. Our estimates retrieved from RDiT are of a compound effect: the causal effect of the 23 March 2020 lockdown announcement on MH service activity and any unobserved sorting/anticipation/adaptation/avoidance effects that may exist but cannot be tested for. The extent to which the results should be interpreted solely as the causal treatment effect of interest depend on the likelihood of other influences. For example, observed increases in mortality are likely to represent effects of the ongoing pandemic rather than lockdown policy, whereas for the service activity measures, adjustment for national mortality (sensitivity analysis 3) is less of an issue and we consider those results to be robust. However, we did not attempt to capture the variation in the timing or magnitude of infection and mortality rates across the UK.

The estimation of mortality increase in our analysis is clearly reflecting a before–after comparison with no population control and is also restricted to all-cause mortality as an outcome with no information available across sites on causes of death responsible. However, changes in all-cause mortality are as valid as those of specific COVID-19 mortality because the former encompasses a wider range of impacts beyond the direct effects of the infection, including adverse consequences of reduced access to medical care as well as suicide mortality as a consequence of reduced MH support. A report from one of the constituent sites indicates that most of the excess mortality in MH service users during March 2020 to June 2020 was from deaths attributed to COVID-19, as well as small excesses in deaths with dementia as an underlying cause.35 In that study, there was also an excess in unexplained deaths compared with previous years, many of which were likely to involve cases awaiting inquests, although these will not necessarily result in suicide verdicts, and estimates to date (for the UK as well as a number of other nations) have not indicated increased suicide rates, at least in the early stages of the pandemic,36 and non-fatal self-harm presentations have also been reported as relatively low during that early period.18 37

Strengths and limitations

Study strengths include the use of relatively ‘real-time’ mental healthcare data from 10 large providers or routinely collected datasets, allowing investigation of changes in service activity following dramatic and rapid transitions at a national scale. The size of the data permitted precise estimates and the relative consistency in the direction of findings supports generalisability. Here, we report on what might be termed a ‘natural experiment’; while individuals were not randomly assigned to experience lockdown or not, assignment is assumed to be quasi-random for observations close to the cut-off . Our regression discontinuity design approach allows valid causal effects to be identified38 and has been previously employed with electronic health record data.39 Limitations include that data were drawn from specific services of interest and do not reflect the full activity of the 10 sites; they were also combined by service type and we did not seek to investigate within-service (team-level) variation. Daily contact numbers were quantified from structured fields in electronic health records and might reflect recording behaviour rather than activity levels (eg, if multiple contacts were subsumed within one entry); also, the dichotomy between f2f and non-f2f contact is a relatively crude one and it neither reflects the quality or depth of assessments being recorded nor the platform that was used (eg, telephone vs video). While we sought to harmonise data extractions as much as was feasible, there were inevitable differences in data availability between sites and likely variation in some measurements (eg, in the completeness of mortality data in clinical systems), although we believe there should be robust within-site consistency across the time periods examined. Some IRR heterogeneity may reflect differences in service provision (eg, acute hospital beds covered by a given liaison service). Clearly, the observation period was limited in duration and there will be a continuing need to collect data to evaluate potential longer-term consequences of both the first and subsequent pandemic waves. Finally, it is important to emphasise that findings here represent quantitative estimates of care provision, rather than care as experienced by patients; there is therefore an important need to evaluate patient satisfaction and perception of care and changes in care delivery before drawing conclusions regarding beneficial or adverse consequences.

Implications

The alterations in MH service activity observed here have been profound and reflect the very limited time period over which to implement responses to the COVID-19 ‘lockdown’ policy. The sharp downturn in inpatient admissions and f2f community contacts are potentially concerning, as is the apparently much longer-term shift in balance from f2f to virtual contacts which remains a marked but unevaluated change in care. On the other hand, as argued above, reduced inpatient admissions and crisis service contact might potentially reflect a reduced demand due to better MH, and virtual contacts might represent an option of care with potential advantages in the future, extending opportunities for patient choice and reducing travel and time costs for both patients and clinical services. Predicting the continuing impact of COVID-19 and related infection-control policies on MH service provision and demand remains challenging. For example, while national ‘shared’ emergencies may be experienced as less isolating for people with mental disorders and some consequences may be less problematic than feared,32 treatment pathways are likely to have been disrupted significantly, particularly since social distancing/lockdown policies were imposed (continuously or intermittently) over many months. It is therefore imperative that national and international collaborations are established to monitor MH service needs and experiences of care as closely as possible, to identify adverse outcomes early enough for rapid intervention—in particular, to understand how people with new and existing severe MH conditions can best be supported in a pandemic, given the need for socially distanced care input.