Diagnostic codes that really raise suspicion of sepsis

Suspicion of sepsis : Improving the Signal to Noise Ratio

The concept of the SoS paper is reasonable (Inada-Kim M, Page B, Maqsood I, et al Defining and measuring

suspicion of sepsis: an analysis of routine data BMJ Open 2017;7:e014885. doi: 10.1136/bmjopen-2016-014885).

We know that coding of sepsis is poor.

This absence of reliable data makes it hard to compare approaches over time and between locations.

The SoS paper basically lists a lot of codes that the authors associated with infection.

The authors identified 267 codes indicating possible infection.

There were 47475 cases with these codes as a primary diagnosis, with 3440 associated deaths.

We felt that several of these reflected conditions that were not primarily infective in nature
(or at least in which antibiotics would not be a main component of acute management)
We removed the most common of these.

11.         N12.X - Tubulo-interstitial nephritis, not specified as acute or chronic

13.         J69.0 - Pneumonitis due to food and vomit

54.         N10.X - Acute tubulo-interstitial nephritis

62.         J84.9 - Interstitial pulmonary disease, unspecified

73.         N71.9 - Inflammatory disease of uterus, unspecified

86.         K57.1 - Diverticular disease of small intestine without perforation or abscess

90.         N48.1 - Balanoposthitis

99.         N48.2 - Other inflammatory disorders of penis

130.      J69.8 - Pneumonitis due to other solids and liquids

The remaining 258 codes had 45521 cases with 3163 associated deaths.

Figure 1 shows the cumulative frequency of these infection diagnoses ranked by coding frequency.

95% of cases are captured using the first 76 diagnostic codes.

We felt that many of the cases reflected infections that are highly unlikely to progress to death
One would expect this with a condition associated with sepsis.
We ranked the remaining codings by number of deaths.
95% of deaths would be captured in just 24 diagnostic codes.

We added in a number of codes that were highly likely to represent relatively common sepsis conditions,
with mortality over 10% and more than 10 cases.

117.      A49.9 - Bacterial infection, unspecified

118.      A40.0 - Sepsis due to streptococcus, group A

121.      A41.2 - Sepsis due to unspecified staphylococcus

131.      A40.3 - Sepsis due to Streptococcus pneumoniae

116.      G00.9 - Bacterial meningitis, unspecified

Altogether, these 29 codes accounted for 29424 episodes with 3026 deaths.

We grouped these conditions into overarching infection codes :

Diagnosis	admissions	deaths	Mortality
Pneumonia	15648	2091	13%
J22.X - Unspecified acute lower respiratory infection	3808	167	4%
J18.9 - Pneumonia, unspecified	3126	483	15%
J44.0 - COPD with acute lower respiratory infection	2854	184	6%
J18.0 - Bronchopneumonia, unspecified	432	242	56%
J13.X - Pneumonia due to Streptococcus pneumoniae	112	11	10%
J15.1 - Pneumonia due to Pseudomonas	51	14	27%
J18.1 - Lobar pneumonia, unspecified	5265	990	19%
UTI	7088	285	4%
N39.0 - Urinary tract infection, site not specified	7088	285	4%
Skin, soft tissue, bone	3021	73	2%
L03.1 - Cellulitis of other parts of limb	2983	62	2%
M72.6 - Necrotizing fasciitis	38	11	29%
Sepsis	2615	426	16%
A41.9 - Sepsis, unspecified	1882	280	15%
A41.5 - Sepsis due to other Gram-negative organisms	360	40	11%
A41.8 - Other specified sepsis	87	32	37%
A41.0 - Sepsis due to Staphylococcus aureus	70	24	34%
A41.1 - Sepsis due to other specified staphylococcus	40	9	23%
A40.8 - Other streptococcal sepsis	34	6	18%
J15.0 - Pneumonia due to Klebsiella pneumoniae	27	8	30%
A40.9 - Streptococcal sepsis, unspecified	21	3	14%
R57.2 - Septic shock	21	11	52%
A49.9 - Bacterial infection, unspecified	19	3	16%
A41.2 - Sepsis due to unspecified staphylococcus	19	4	21%
A40.0 - Sepsis due to streptococcus, group A	19	4	21%
A40.3 - Sepsis due to Streptococcus pneumoniae	16	2	13%
Abdominal	723	122	17%
K57.2 - Diverticular disease of large intestine with perforation and abscess	231	20	9%
A04.7 - Enterocolitis due to Clostridium difficile	208	20	10%
K63.1 - Perforation of intestine (nontraumatic)	136	51	38%
K65.0 - Acute peritonitis	96	13	14%
K65.9 - Peritonitis, unspecified	52	18	35%
Biliary	240	18	8%
K83.0 - Cholangitis	240	18	8%
Endocarditis	89	11	12%
I33.0 - Acute and subacute infective endocarditis	89	11	12%
Grand Total	29424	3026	10%

In conclusion, we captured 96% of infection deaths using just 29 diagnostic codes.
It is thus possible to simplify the analysis of Inada Kim et al considerably,
and by so doing improve the signal to noise ratio of the analysis.
We still included some conditions that are not primarily infective (eg abdominal perforation)
However, this led us to consider that a practical definition of sepsis might be
"A diagnosis for which antibiotics would generally be seen as part of acute management"

The paper only looks at primary codes, and this might miss hospital acquired sepsis.
We agree that this is likely to increase noise, but this is an area that merits further research.

We also note that the authors state “although there are non-bacterial causes of sepsis (eg, viruses, protozoa), these are (generally) far less common and amenable to treatment” 
Given that influenza is one of the commonest causes of death from infection, this is perhaps hard to justify.

This will help comparative studies of sepsis management,
and provide focus on the areas that are likely to be benefit from attention.