Building Your First Hospital Antibiogram: A Step-by-Step Method

Start by fixing the period. Twelve consecutive months is the standard unit, and shorter periods should only be used when you have deliberately chosen a narrower question and have the isolate volume to support it. Then decide the specimen scope. Include clinically significant diagnostic isolates. Exclude surveillance and screening specimens, because a methicillin-resistant Staphylococcus aureus nasal screening programme will flood the Staphylococcus line with resistant isolates that were never infections and will make your empirical guidance for skin and soft tissue infection wrong.

Exclude environmental and staff-screening isolates for the same reason. Handle contaminants explicitly: coagulase-negative staphylococci from a single blood culture bottle are usually contamination and, if included, will distort the blood-isolate table badly. Write down the rule your laboratory will apply, apply it consistently, and state it on the published antibiogram so a clinician can see what was counted. An antibiogram with an unstated inclusion rule cannot be compared with anything, including its own previous edition.

Only verified final results should count. Isolates with intermediate results need a stated convention, and the usual one is to report percentage susceptible only, excluding intermediate from the numerator while keeping it in the denominator. Suppressed results, where the laboratory tested an agent but did not release it on the patient report for stewardship reasons, should still enter the antibiogram, because the antibiogram is a population artefact and the suppression rule was a reporting decision for an individual patient.

The single most important rule is deduplication: count the first isolate of a given species per patient per analysis period, regardless of how many times that species was subsequently isolated from that patient. Without it, one ICU patient who grows Klebsiella pneumoniae from six specimens over three weeks contributes six times the weight of a patient who grows it once, and that patient is by definition the one with the most resistant organism. Skip deduplication and your table will report resistance for a small group of very sick patients as though it applied to everyone.

There are more refined variants. Some laboratories deduplicate per patient per species per specimen source, so a urinary isolate and a blood isolate of the same species from the same patient both count. Others deduplicate within a shorter window, such as thirty days, so a genuinely new infection three months later is counted again. Both are defensible. What is not defensible is choosing a variant silently, or changing it between editions, because either makes your year-on-year trend meaningless.

Deduplication has a real cost: it shrinks your isolate counts, sometimes dramatically, and it is the reason a first antibiogram often falls below reportable thresholds for organisms the hospital sees regularly. That is not a reason to abandon the rule. It is information about the size of your dataset, and the correct response is to lengthen the period or aggregate at a broader level rather than to inflate the counts by counting the same patient repeatedly.

CLSI M39 advises against reporting a percentage susceptibility for an organism-antimicrobial combination with fewer than thirty isolates, and the reasoning is straightforward. With twelve isolates, a single result flips the percentage by more than eight points, and with five isolates the figure carries no useful precision at all. Publishing 60 per cent susceptibility from five isolates gives a clinician a number that looks like evidence and behaves like noise. The threshold is not bureaucratic caution; it is the point below which the table stops being informative.

Below the threshold you have three honest options. Report the raw counts instead of a percentage, so a reader sees three of five rather than 60 per cent. Aggregate two or three years of data for that organism and label the period clearly. Or suppress the row entirely and state in the footnotes that the organism was isolated but fell below the reporting threshold. What you must not do is publish the percentage with a small asterisk that nobody reads. Clinicians read tables, not footnotes.

Small hospitals hit this constantly, and it is worth being explicit about the consequence: a 150-bed hospital may have only enough Pseudomonas aeruginosa or Acinetobacter isolates to report at hospital level, never at unit level. That is a genuine limit on what your own data can tell you, and it is the honest reason to lean on ICMR antimicrobial resistance surveillance network reports and state-level data for those organisms while your own dataset accumulates.

WHONET, the free software maintained under WHO auspices for microbiology laboratory data management, is the pragmatic tool for a hospital doing this for the first time. It handles the parts that are tedious in a spreadsheet: importing from a laboratory information system through BacLink, applying first-isolate deduplication as a built-in analysis option, applying breakpoint interpretation, and generating the susceptibility summary directly. It is not elegant software and the learning curve is real, but it encodes the methodology, which a spreadsheet built by a keen registrar does not.

The work that actually consumes time is data mapping, not analysis. Organism names, antimicrobial codes, specimen types and ward identifiers all have to map from your LIS vocabulary to WHONET's, and the mapping has to be maintained when the laboratory adds a test or the hospital renames a ward. Budget a few days of a microbiologist's and an IT person's time for the first configuration, then a short review before each annual run. Most failed antibiogram projects fail here, in unglamorous data plumbing, not in statistics.

The alternative is to have your LIS or hospital system produce the extract to a defined specification and analyse it directly. This is reasonable if your vendor will implement first-isolate deduplication correctly and expose the inclusion rules as configuration rather than burying them in code. Ask to see how deduplication is implemented before you accept it. A report that silently counts every isolate is worse than no report, because it carries the hospital's letterhead and nobody will question it.

A hospital-wide antibiogram is the wrong instrument for choosing empirical therapy in an ICU, because ICU flora and ICU resistance are systematically different from the wards. Where isolate volumes permit, a separate ICU antibiogram is the highest-value split you can make. The second most useful split is by specimen type: a urinary antibiogram covering Escherichia coli and Klebsiella species from urine will guide far more prescriptions in a general hospital than any other single table.

Splitting has a hard arithmetic limit. Every split divides your isolate counts, and the thirty-isolate threshold applies to each cell of each sub-table, not to the parent. A hospital that splits by unit, by specimen and by inpatient versus outpatient will produce a document that is mostly suppressed rows. Decide the splits by what will change a prescribing decision, then check whether the counts survive. If they do not, publish fewer tables and say why, rather than publishing thinner ones.

There is also an interpretation trap specific to unit-level tables. An ICU antibiogram reflects the organisms cultured in the ICU, which includes organisms the patient brought in from a ward or another hospital. It is not a measure of ICU-acquired resistance and should never be read as one. Label each table with what it contains, in one line, at the top. Clinicians who misread a table are usually responding to an ambiguous heading rather than misunderstanding microbiology.

Annual publication is the standard cadence, and it is right for the antibiogram itself. It is not right for the awareness it should generate. A table published once a year and emailed as a PDF attachment will be opened by the people who already care. Pair the annual publication with a short circulation of what changed from last year, framed as prescribing implications rather than percentages: the empirical choice for complicated urinary tract infection has changed, and here is the sentence to remember.

Physical placement beats digital distribution in most Indian hospitals. A laminated single-page version in the ICU, the emergency department and the doctors' room gets consulted. Where prescribing happens on screen, the relevant susceptibility summary should surface at the point of ordering rather than in a separate document; systems such as HealUDoc can hold the current empirical guidance against the order set so the clinician is not asked to remember which edition is current.

Version control is the unglamorous requirement that assessors ask about. Each edition should carry a version number, the analysis period, the inclusion and deduplication rules used, the number of isolates behind each organism row, and the date the medical committee approved it. When the antibiotic policy is revised in response to the antibiogram, the policy should cite the antibiogram version it was built on. That chain of documents is what converts a laboratory report into an accountable clinical governance artefact.