The analysis was conducted by the Centre for Behavioural and Implementation Science Interventions (BISI) at the Yong Loo Lin School of Medicine, National University of Singapore (NUS Medicine), working alongside the clinical leadership of NUHS@Home. Rather than simply reporting outcomes, the team applied implementation science methods, a discipline concerned with how evidence-based practices can be embedded reliably into real-world systems and workflows, to understand what it actually takes to move a small pilot into a full-scale service. Professor Nick Sevdalis, Director of BISI, worked with Dr Stephanie Ko, lead of NUHS@Home, to systematically map the considerations and approaches needed for scaling the programme between 2022 and 2025. The findings were presented at the Behavioural and Implementation Sciences in Health, Asia Conference 2026 (BISHAC 2026), Singapore’s first conference dedicated to these fields, as a real-world case study in accelerating healthcare transformation.

The numbers behind the scale-up are striking. NUHS@Home is NUHS’ Mobile Inpatient Care service, delivering acute-level care to patients who would otherwise require ward-based hospitalisation but who are clinically stable enough to be managed in their own homes. Launched in September 2020 as a three-bed proof-of-concept pilot, the service was subsequently pivoted to support COVID-19 care as part of the national pandemic response, a period that gave the team early operational experience in caring for acutely unwell patients outside hospital walls. Since May 2022, the service has grown from an average of 48 patients a month to more than 300 a month by 2025, a roughly seven-fold increase in patient volume. Virtual beds scaled even faster, expanding fifteen-fold from 6 to 90 over the same period.

Perhaps the most telling statistic concerns the service’s share of overall capacity. At full scale, NUHS@Home beds make up 3.3 percent of total bed capacity across NUHS, up from 0.6 percent at the start of the scale-up, a roughly 5.5-fold increase in relative share. That figure captures a structural shift rather than a temporary surge: a growing proportion of hospital-level care within the cluster is now being delivered safely at home instead of in a ward. Over the scale-up period, the programme saved an estimated 21,800 hospital bed days, capacity that was freed up for patients whose conditions genuinely require ward-based treatment. In systems where bed occupancy routinely runs close to operational limits, the ability to create that much headroom without constructing a single new ward carries obvious strategic value.

Crucially, the analysis found that rapid growth did not come at the cost of quality. As the service expanded, 30-day readmission rates and home visit frequency showed no statistically significant difference across the scale-up period. Readmission rates are one of the most closely scrutinised safety indicators for hospital-at-home models, because a deterioration at home that ends in an unplanned return to hospital is the key failure mode the model must guard against. Home visit frequency, meanwhile, is a proxy for care intensity, since in-person assessment by clinical teams is what distinguishes true hospital-at-home care from remote monitoring alone. Holding both steady while monthly patient volumes grew seven-fold suggests that the expansion was achieved by adding capacity and broadening eligibility rather than by thinning the care delivered to each patient.

The implementation science analysis identified three factors associated with successful scale-up. The first was growing the healthcare workforce trained to deliver hospital care at home, which is a non-trivial challenge: clinicians accustomed to ward-based practice need to learn to work autonomously in patients’ homes, coordinate remotely, and make decisions without the immediate backup of a physical hospital environment. The second was expanding the range of clinical conditions that can be managed at home, including patients with cancer and those recovering after surgery, groups that push the model well beyond the simpler infectious-disease and general-medicine cases with which many hospital-at-home programmes begin. The third was strengthening quality and safety processes, ensuring that governance, escalation pathways and monitoring kept pace as the service multiplied in size and complexity.

For the clinical leadership, the human dimension of those statistics is the point of the exercise. “Behind every number in this study is a patient who got to recover in their own bed, surrounded by family, instead of in a hospital ward. Working with BISI helped us understand how to scale that option safely, so more patients now have the choice of receiving hospital-level care where they’re most comfortable,” said Dr Stephanie Ko, Lead of NUHS@Home. The sentiment reflects a broader body of patient-experience research suggesting that familiar surroundings, better sleep and reduced exposure to hospital-acquired infection are among the advantages patients cite when hospital-level care can be delivered at home.

The work also signals an institutional commitment to making this kind of translation routine rather than exceptional. To accelerate the movement of research into frontline care across NUHS, a dedicated BISI team will support the scale-up of effective projects under the auspices of the newly established NUHS Health Services and Population Health (HSPH) Unit, which combines behavioural and implementation sciences with data science and health economics. Established on 1 August 2026, the HSPH Unit is positioned to drive health services innovation and population health improvement across the cluster. “Good ideas don’t scale by themselves. With the right method, a proven innovation can be grown quickly, safely and at lower cost. By applying implementation science methods, we helped NUHS@Home identify how to best grow and get the right care to more patients faster. With the new NUHS Health Services and Population Health Unit, our goal is not only to help scale more effective programmes in NUHS, but to help build the cluster’s capability to translate evidence into better care, routinely,” said Professor Nick Sevdalis, Director of BISI and the HSPH Unit.

The NUHS@Home findings were presented at BISHAC 2026, held from 3 to 5 September 2026 and convened by BISI, one of the few centres in the world to combine behavioural science and implementation science expertise within a single organisation. The distinction between the two fields matters: behavioural science explains how people make decisions, while implementation science explains how evidence-based practices can be embedded into systems and workflows. BISHAC 2026 marks the first time the two fields have been brought together in a dedicated Singapore conference, with a deliberate focus on Asian health systems, where ageing populations, rising chronic disease burden and constrained infrastructure make scalable alternatives to conventional ward care especially urgent.

The conference also served as the launchpad for a pan-Asian research collaboration aimed at closing the persistent gap between what health research shows and what patients actually receive. The Hub for Implementation, Voices and Engagement (HIVE) initiative, jointly led by BISI and the School of Public Health of Zhejiang University, brought together implementation science researchers from across the region to set shared priorities for the next five to seven years. In total, 67 senior scientists across 15 Asian countries contributed expert views and identified 27 priority areas for accelerating the translation of evidence into practice. Among the top priorities were advocacy for implementation work with research funders, close working with communities, partnerships between researchers and policy makers, regional mentorship and training programmes, and sustainable financing for scaling successful health interventions. For the NUHS@Home team, the message of both the service data and the conference is the same: the evidence that hospital-level care can be delivered at home is no longer in doubt, and the frontier now lies in the disciplined, measurable work of helping proven models grow quickly, safely and affordably across diverse health systems.

Subject of Research: Implementation science-guided scale-up of a hospital-at-home service in Singapore

Article Title: NUHS@Home scales bed capacity 15-fold, saving 21,800 hospital bed-days

Article References: NUHS@Home scales bed capacity 15-fold, saving 21,800 hospital bed-days. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: hospital-at-home, NUHS@Home, implementation science, virtual beds, bed capacity, patient safety, readmission rates, NUS Medicine, BISHAC 2026, healthcare scaling, Mobile Inpatient Care, population health

The study, published in BMC Geriatrics, analysed 8,208 ambulance assignments involving adults aged 65 and older in the Stockholm region between 2017 and 2019. All of these patients met strict clinical criteria for direct prehospital conveyance to a geriatric unit: they had urgent but non-life-threatening conditions, vital signs within normal reference ranges, and one of twelve predefined conditions considered suitable for geriatric assessment outside the emergency department. Despite this careful standardisation of patient eligibility, only 603 patients, roughly 7 percent of the sample, were actually conveyed directly to a geriatric unit. The remainder were taken to emergency departments, the default destination for most ambulance transports.

The headline finding is the sheer scale of geographic variation. Across the eight hospital service areas included in the analysis, each anchored by a distinct geriatric unit, crude conveyance rates ranged from 4 percent to 21 percent. In other words, a clinically similar older adult was more than five times as likely to be taken straight to geriatric care in some parts of Stockholm as in others. To disentangle what might be driving this pattern, the researchers employed multilevel logistic regression, a statistical framework designed for data with a nested structure, in this case patients clustered within hospital service areas. Random intercepts for each area allowed the probability of direct admission to vary geographically, while a suite of measures quantified the magnitude of that variation.

The technical vocabulary of the paper rewards attention. The intraclass correlation coefficient, or ICC, estimated that around 10 percent of the total variation in conveyance outcomes was attributable to differences between areas in the unadjusted model. The median odds ratio, or MOR, translates area-level variance into a more intuitive quantity: in the fully adjusted model, an MOR of 1.54 means that two otherwise identical patients living in different hospital service areas could face odds of direct geriatric admission differing by a factor of 1.54, simply as a function of residence. The proportional change in variance, or PCV, tracked how much of the between-area variation disappeared as successive covariates were added to the models.

Crucially, the researchers separated potential explanations into demand-side factors, which reflect patient characteristics and healthcare need, and supply-side factors, which reflect the organisation and capacity of the care system itself. The demand-side variables included age, sex, socioeconomic status, living conditions, comorbidity burden, and previous hospitalisation. The results were striking: adjusting for these patient characteristics did not shrink the geographic variation at all. In fact, the between-area variance increased slightly, by nearly 9 percent, suggesting that differences in patient mix between areas actually masked rather than explained the true extent of the disparity. Where older people live matters for their access to geriatric care, and it is not because they are sicker or different in ways the registers can capture.

Supply-side factors told a different story. Three variables stood out as correlates of the observed heterogeneity. Geographic proximity to a geriatric unit reduced the estimated between-area variance by 26 percent, consistent with a well-documented tendency of ambulance crews to convey patients to the nearest suitable facility when distance differentials widen. Ownership of the geriatric unit, whether public or private, was associated with a 27 percent reduction in variance, although this finding proved fragile under alternative statistical approaches. Most powerfully, the number of staffed geriatric beds in an area reduced the between-area variance by 38 percent, implying that capacity itself, the sheer availability of specialised beds, shapes who gains access to this form of care. In the fully adjusted model combining all demand- and supply-side factors, the between-area variance fell by 41 percent, yet a substantial residual disparity remained.

The study also examined timing, finding that conveyance during office hours was significantly associated with direct geriatric admission, even though it did not explain the differences between areas. This pattern echoes previous research suggesting that out-of-hours conveyance rates drop when diagnostic support, laboratory access, and staffing are reduced, and when ambulance personnel feel heightened responsibility for patient safety. An incorrect decision not to convey an older person could lead to adverse health events, while a defensive decision to send everyone to the emergency department wastes resources and exposes frail patients to environments poorly suited to their needs. Emergency departments, designed for rapid assessment of acute single conditions, have been shown in prior studies to place older adults at risk of long waits, fragmented care, and functional decline, particularly for those with multimorbidity, frailty, or geriatric syndromes such as falls and delirium.

The stakes of this routing question are underscored by an accumulating evidence base on geriatric units themselves. Comprehensive geriatric assessment delivered by interdisciplinary teams has been associated with improved functional status and reduced institutionalisation compared with usual care pathways. Research from France has suggested that direct admission to acute geriatric units, bypassing the emergency department, may shorten length of stay, reduce post-acute transfers, and even prove cost-effective for selected patients. In Stockholm, a prehospital decision support system in use since 2011 directs older patients with specific conditions and mild acute illnesses toward alternative care pathways, including direct conveyance to geriatric units, and patients receiving such direct care have expressed satisfaction with it, typically remaining at the geriatric unit for their entire care episode.

The authors are careful about the limits of their findings. With only eight hospital service areas, variance estimates in multilevel logistic models can be biased and imprecise, and the team tested this vulnerability with three robustness analyses, including a Bayesian multilevel model with weakly informative priors. That Bayesian analysis produced a posterior mean variance roughly twice that of the primary frequentist estimate, with a wide highest posterior density interval, indicating substantial uncertainty around the magnitude of the geographic variation and around the specific contribution of unit ownership. The authors also note that the study lacked measures of organisational factors such as real-time bed availability, laboratory capacity, and diversion practices, as well as data on cognitive impairment and functional status, any of which could contribute to the unexplained residual variation. Distance was calculated from residential addresses rather than actual pickup locations, and the study measured conveyance rather than patient outcomes.

Even with these caveats, the implications are difficult to ignore. In a universal, tax-funded healthcare system like Sweden’s, where all accredited providers operate under the same regulatory and reimbursement rules, one might expect access to specialised care pathways to be relatively uniform. Instead, the study suggests that the geography of geriatric bed capacity and the location of units themselves may quietly ration access to a form of care that clinical evidence favours for frail older adults. The authors point toward practical levers: reviewing why non-conveyance to geriatric units rises outside office hours, and considering how the placement and capacity of geriatric services might be planned to support more consistent and equitable pathways. At the same time, they caution against naive capacity expansion, since supply-sensitive utilisation is a well-known phenomenon in health services research. The deeper message is that for older adults, the emergency care pathway begins not at the hospital door but in the ambulance, and the organisation of what lies downstream of that first decision deserves the same scrutiny as any hospital intervention.

Subject of Research: Geographic variation in direct ambulance conveyance of older adults to geriatric units versus emergency departments in Stockholm, Sweden

Article Title: Disparities in direct access to acute geriatric care – an observational analysis of geographic variations of ambulance conveyance

Article References: Varg, S., Vicente, V., Hagman, M., Johansson, N., Lindgren, P., Heintz, E., & Rehnberg, C. (2026). Disparities in direct access to acute geriatric care – an observational analysis of geographic variations of ambulance conveyance. BMC Geriatrics, 26(1), Article 1244. https://doi.org/10.1186/s12877-026-08370-1

Image Credits: AI Generated

DOI: 10.1186/s12877-026-08370-1

Keywords: geriatric care, ambulance services, emergency department, geographic variation, older adults, prehospital care, hospital service areas, health equity, multilevel modelling, Sweden, emergency medical services, bed capacity