In October 2019, a few months before anyone had heard of SARS-CoV-2, the Global Health Security Index ranked the United States as the best-prepared country in the world for a pandemic, with a score of 83.5 out of 100. The United Kingdom came second.
We know how that went.
The index was not built carelessly. It was a serious attempt, by serious people, to score countries on detection, response, health system capacity and more. Its failure is instructive precisely because it was so thorough. It measured what countries had. It did not measure what they would do with it under pressure.
That gap is where most resilience measurement still lives.
Assets are not the same as capacity
Most resilience frameworks are, at heart, inventories. Beds per thousand people. ICU capacity. Laboratory networks. The existence of a national preparedness plan. Stockpiles of personal protective equipment.
These are easy to count and easy to compare across countries. They are also poor predictors of how a system behaves when it is hit. A 2022 study in Frontiers in Epidemiology looking at major European countries found that in mid-2020, higher GHS Index scores were actually associated with higher COVID-19 death rates per capita. The relationship only flipped to the expected direction by July 2021, well into the vaccine era.
A large analysis in The Lancet in 2022, covering 177 countries and territories, went further. It found that the GHS Index and similar preparedness measures were not associated with infection or infection-fatality rates. Trust, in government and between people, was associated with fewer infections.
Trust does not appear on most resilience dashboards. It is awkward to measure and impossible to buy quickly. That is exactly why it matters.
The uncomfortable incentive behind inventory metrics
Why do we keep measuring assets? Partly because they are measurable. But there is a less flattering reason.
Asset metrics are fundable. A ministry can ask a donor or a finance ministry for more ventilators, a new laboratory or an emergency operations centre, and the purchase shows up as progress on the next scorecard. Nobody gets a ribbon-cutting for a procurement system that can switch suppliers in a week, or for a workforce that has not been burned out by a decade of vacancies.
So the indicators drift toward things that can be bought and photographed, and systems get optimised for the scorecard rather than for the shock. A plan that sits on a shelf scores the same as a plan that has been rehearsed three times with the people who will actually run it.
Metrics that tell you something
If the question is whether a system will keep functioning when something goes wrong, the better indicators tend to measure behaviour, not holdings. A few worth taking seriously:
- Continuity of essential services during disruption. WHO’s first pulse survey in 2020 found that 90% of the 105 responding countries reported disruptions to essential health services. The useful metric is not whether disruption happened, since it did almost everywhere, but how deep it went and how quickly routine immunisation, antenatal care and chronic disease management recovered.
- Time to recover, not just time to respond. How long after a flood, cyberattack or outbreak does a district’s service volume return to baseline? This one needs routine data that most countries already collect.
- Surge that is real, not theoretical. Beds are only capacity if there are staff to run them. A staffed-bed measure, or the number of qualified people who could be redeployed within a week, is harder to inflate.
- Supply chain depth. How many weeks of stock for a basket of essential medicines, and how many suppliers for each? Single-source dependencies are where shocks travel fastest.
- Excess mortality after the fact. WHO estimated that the pandemic was associated with about 14.9 million excess deaths in 2020 and 2021. Excess mortality is blunt, but it is honest. It captures deaths from the disease and from the care that did not happen.
None of these is exotic. Most can be built from routine health information systems. What they share is that they measure the system in motion.
The case for fewer, uglier indicators
The instinct in global health is to build a composite index with dozens of sub-indicators and a single headline number. Composites are useful for advocacy. They are poor tools for management, because a country can improve its score by adding assets it will never use while its real weakness stays hidden in a sub-indicator nobody reads.
A resilience measure a ministry can actually manage against is probably short, local and slightly embarrassing. Recovery time for immunisation coverage after the last disruption. Proportion of district facilities that ran out of a tracer medicine in the past quarter. Nurse vacancy rate in the hardest-to-staff region. These do not rank countries neatly. They do tell a health officer where the next shock will hurt.
There is also a case for stress tests: tabletop exercises and simulated disruptions, scored on how quickly decisions actually get made. Banks were made to do this after 2008. Health systems mostly still grade themselves on whether a plan exists.
The mechanism: why inventories mislead
It is worth being precise about why asset counts fail, because the reason is not simply that they are incomplete.
A health system under stress behaves like a chain of dependencies. A ventilator is only useful if there is oxygen, a trained nurse, a working power supply, a drug to sedate the patient and a physician who knows when to wean. An intensive care bed is only useful if the hospital can get the patient there, which depends on ambulances, referral rules and communication between facilities. A laboratory is only useful if samples reach it and results reach someone who acts on them.
Inventory metrics count the links. Resilience depends on the weakest one. A country can double its ventilators and add nothing to real capacity if the bottleneck is oxygen or nurses. Worse, the extra equipment can create a false sense of security that delays attention to the actual constraint.
This is also why the same assets can produce very different results in different countries. Two systems with identical bed counts can respond completely differently if one has clear command structures, trusted public communication and staff who can be redeployed quickly, and the other has fragmented authority and an exhausted workforce. Those differences are organisational and social. They do not appear in inventories, and they are exactly what the Lancet analysis picked up through trust.
A worked scenario: two districts, one flood
Imagine two neighbouring districts hit by the same severe flood.
On paper, District A looks stronger. It has more hospital beds per head, a newer district hospital and a larger stock of emergency supplies. District B has an older hospital and fewer beds, but it has held two emergency simulations in the past three years, maintains an up-to-date list of community health workers with phone numbers, and has an agreement with a neighbouring district to share ambulances and staff.
When the flood arrives, District A’s new hospital sits on low ground and loses power. The emergency supplies are in a warehouse that cannot be reached. Staff do not know who is in charge of the response, and decisions wait for instructions from the capital.
District B’s hospital also struggles, but community health workers are contacted within hours and asked to identify pregnant women, dialysis patients and people on oxygen. Ambulances from the neighbouring district arrive under the existing agreement. Routine immunisation resumes within weeks because someone was assigned to restart it.
A national resilience scorecard compiled the year before would probably have ranked District A higher. A measure of recovery time afterwards would tell a very different story. That is the whole argument in one example.
The strongest counterargument
Defenders of asset-based indices make a reasonable case. Assets are necessary, even if not sufficient. A country with no intensive care capacity, no laboratory network and no stockpile cannot be resilient however well organised it is. For many low-income countries, the most urgent problem really is a shortage of basic physical and human resources. Behavioural metrics are also harder to standardise, easier to game in their own ways and more expensive to collect.
This argument partly holds. Assets set a ceiling on what any system can do, and in the poorest settings that ceiling is often the binding constraint. Composite indices also serve a genuine purpose in advocacy and in directing international finance.
But the pandemic showed that above a basic threshold, assets stop predicting outcomes well. Among middle- and high-income countries in particular, the gap between capacity on paper and performance in practice was large. The lesson is not to abandon asset measures. It is to stop treating them as a proxy for resilience once basic capacity exists, and to add measures of how the system actually behaves.
What this means for different players
Ministries of health could pick a small number of behavioural indicators, such as recovery time for routine services after disruption and stock-out rates for tracer medicines, and report them publicly alongside asset measures. Choosing indicators that are already in routine data keeps the cost down.
Hospital leaders should measure how quickly their own services recover after an incident, whether a power cut, an IT failure or a surge. Hospitals that do not know their recovery time for elective surgery or outpatient clinics after a disruption do not really know how resilient they are.
Funders should be wary of financing assets without the operating costs and organisational investment needed to use them. Equipment donated into a system that cannot staff or maintain it inflates the scorecard and adds little resilience. Funding exercises, supply chain diversification and workforce retention is less photogenic but often more effective.
Clinicians are often the first to notice when a system is running on its reserves: rising vacancies, informal workarounds, staff covering for missing colleagues. Creating channels for that frontline intelligence to reach planners is a cheap resilience measure in itself.
What to watch next
Several developments will show whether resilience measurement is maturing. One is whether updated preparedness frameworks give more weight to governance, trust and service continuity than earlier indices did. Another is whether countries begin routinely publishing recovery-time data after disruptions, not just incident reports. A third is whether donors start tying preparedness financing to demonstrated performance in exercises rather than to the delivery of equipment.
Each of these would be a sign that the field has absorbed the lesson of 2020. None of them is guaranteed.
Where this leaves us
Resilience is not a stock of things. It is a property of how a system behaves under stress, and behaviour only shows up in behavioural data. The countries that performed well in the pandemic did not all have the most beds or the best scores. Many of them had functioning public health relationships, clear lines of authority and populations willing to follow advice.
The next ranking will be published, and some country will again top it. The more useful question is whether it measures anything that would have predicted the last crisis.
A resilience metric that cannot be falsified by a real shock is not measuring resilience. It is measuring the budget.
Which resilience indicator has actually helped your system make a decision during a disruption, and which ones only showed up in reports? Tell us in the comments.
