HAI Benchmarking for Indian Hospitals: Comparing Rates Honestly

Four sources come up most often. The Indian Council of Medical Research runs a healthcare-associated infection surveillance network across participating tertiary hospitals and publishes periodic reports of device-associated infection rates by unit type. The International Nosocomial Infection Control Consortium publishes multinational data drawn largely from limited-resource settings, including many Indian sites. The US National Healthcare Safety Network publishes extensive American data. The European Centre for Disease Prevention and Control maintains its own network for European hospitals.

These do not measure the same thing. The definitions differ in detail, the participating hospitals are self-selected, and crucially the settings differ in how aggressively they look. NHSN data comes from a system with mature electronic surveillance and high blood culture rates. INICC data has consistently described higher device-associated rates in limited-resource settings than the American figures, and the authors themselves attribute part of that to real differences in practice and part to differences in the systems being compared. Read the methods section of any report before you read its tables.

There is also a participation bias that nobody can adjust away. Hospitals that join a voluntary surveillance network are hospitals that have an infection control programme good enough to join one. Their published rates are not a national average; they are the rates of the better-organised end of the distribution. Comparing yourself against them is closer to comparing yourself against a peer group you have not yet been admitted to than against a typical hospital.

The most powerful hidden variable in any infection rate is how hard the hospital looked. A unit that draws blood cultures from every febrile patient will identify bloodstream infections that a unit which starts empirical antibiotics without culturing will never see. Both units may deliver identical care. Only one has a numerator. This is why a low reported HAI rate in a hospital with low culture rates should be read as a data quality finding rather than a clinical achievement.

You can measure your own detection intensity, and you should, because it makes your trend interpretable. Blood cultures per 1000 patient days, urine cultures per 1000 catheter days, and the proportion of ventilated patients with a respiratory sample sent are all straightforward to derive from laboratory data. Track them alongside the infection rates on the same report. When your CLABSI rate rises in a quarter where blood culture rates also rose sharply, you have a plausible explanation for at least part of the movement.

Case-finding method matters as much as culture volume. A dedicated infection control nurse reviewing every positive blood culture daily will find more than a system relying on ward staff to notify. Adding an ICN, or moving from paper notification to an automated laboratory alert, will raise your measured rate without any change in what happened to patients. Log every such change on the trend chart, because otherwise your own improvement in surveillance will be read by a future committee as a deterioration in care.

Case mix is the obvious one and the least often quantified. A unit taking transplant recipients, prolonged-ventilation cases and referrals arriving with established sepsis carries a fundamentally different infection risk from a post-operative step-down unit that keeps patients for forty-eight hours. Published network figures are usually stratified by unit type, which helps, but a medical-surgical ICU in a Delhi referral centre and a medical-surgical ICU in a district private hospital share a label and very little else. Severity scoring, if you already collect APACHE or SOFA data, gives you at least a describable position.

Definitions break comparison more quietly. VAP and ventilator-associated events are different constructs and produce different counts from the same patients. Laboratory-confirmed bloodstream infection and catheter-related bloodstream infection, which requires paired cultures or tip culture evidence, are not interchangeable. Secondary bloodstream infection exclusions vary. Before comparing anything, write down the exact definition version you use and the exact version the comparator used. If you cannot determine the comparator's version, you cannot use it.

Denominator practice is the third. Whether a hospital counts device days by daily census or estimates them from bed occupancy, whether it counts per patient or per device, and whether it includes step-down beds all shift the rate by amounts comparable to the differences people are trying to interpret. Two hospitals with identical care and identical detection can report rates differing by half simply through denominator convention. None of this is visible in a published table.

The standardised infection ratio is the methodological answer to case mix. Instead of comparing a raw rate, you compare observed infections against the number predicted for a hospital with your particular unit types, bed sizes, medical school affiliation and other risk factors, using a regression model fitted to the reference population. A ratio above one means more infections than predicted. It is a genuine improvement on raw rate comparison and it is what mature national programmes report.

It also cannot be borrowed. The predictive model is fitted on the reference network's own data, and applying an American model to an Indian hospital assumes the relationships between risk factors and infection are the same in both settings, which is exactly the assumption in question. Until an Indian network publishes a risk model fitted on Indian data with sufficient coverage, an SIR calculated against a foreign model is a precise-looking number resting on an unexamined assumption.

What you can borrow is the logic. Build your own internal standardisation by splitting your data along the dimensions you know drive risk in your hospital, then compare each stratum only against itself over time. Your ICU against your ICU last year. Your ventilated patients against your ventilated patients. It is less sophisticated than a fitted model and it does not let you claim a position against a national figure, but every comparison it produces is one you can defend line by line.

An internal baseline needs three things: a long enough series, a frozen methodology, and annotation. Twelve consecutive months is the minimum before you have any sense of your own normal variation, and twenty-four is better. During that period the case definitions, the denominator rule, the census time and the case-finding method must not change. If they must change, the series restarts and you say so on the chart rather than splicing the old and new together.

Plot it as a run chart or a statistical process control chart rather than a bar comparison against last month. The question a control chart answers is the only question worth asking of a monthly infection rate: is this month's value within the range this process normally produces, or is it a signal? Two events instead of one in a small ICU is almost never a signal. Eight consecutive months above the median is, even if no single month looked alarming.

Set improvement targets against your own baseline, expressed as a sustained shift rather than a threshold. A target of reducing the median CLABSI rate by a stated proportion and holding it for six consecutive months is measurable, honest and achievable. A target of getting below a published national figure may be unachievable for reasons that have nothing to do with your care, and missing it repeatedly does real damage to the credibility of the quality programme inside the hospital.

Boards want a single number and a comparison, and giving them one uncritically is how bad decisions get made. The better format is a short standard page: your rate, your trend with control limits, your detection intensity, and one line stating explicitly what external comparison is and is not possible. Boards accept this readily when it is presented as rigour rather than evasion, and it protects the quality team from being asked to explain a gap that is an artefact.

Payers, corporate clients and empanelment processes increasingly ask for infection rates, and the honest answer includes methodology. Submit the rate with the definition set, the period and the denominator method attached. A hospital that submits a slightly higher rate with a clear method reads as credible; one that submits a suspiciously low rate with no method invites a closer look. Insurers comparing bids have started to notice which is which.

For a NABH assessment, the expectation is not that your rate is low. It is that the indicator is defined, collected consistently, analysed, compared over time, and acted upon where analysis showed something. An assessor will ask to see the definition, the raw collection sheets, the trend, the minutes where the trend was discussed, and the action that followed. A hospital with a moderate rate and that full chain in place is in a far stronger position than one with an excellent rate and no evidence of how it was produced.