Article Text
Statistics from Altmetric.com
Various methods have been used to evaluate the quality of cause of death statistics.1,2 Traditionally, necropsy findings were deemed as the gold standard to evaluate the accuracy of cause of death certification. However, because of the biased selection of necropsy cases and the decreasing necropsy rate, fewer and fewer evaluation studies have used necropsy findings as the standard.3 Another commonly used standard to evaluate the quality of death certification is the consensus of a panel of physicians reviewing all available information related to the deceased.4 Most of the studies using this method were the byproducts of large cohort studies or randomised clinical trials. These studies wanted to assure that the end point was not biased. The shortcomings of using physician review as the standard were time consuming, costly, not applicable in a large scale and routinely.5,6
As more and more disease specific registries and hospital medical records were computerised, more and more investigators began to use these datasets as the standard to evaluate the quality of cause of death statistics. The merits of this method were time saving, less costly, applicable in large scale and routinely. Several population based studies used computer linkage of cause of death file and hospital discharge file to compare the underlying cause of death (UC) and discharge diagnosis.7–10 Almost “every” deceased were medically attended in developed countries, these studies could thus evaluate the quality of “every” death certification. Johansson and Westerling attempted to develop a systematic and routine monitoring mechanism at the national level that can detect the poor death certification “before” the publication of mortality data.10 Unlike the previous post hoc studies, through this monitoring system, once the poor quality of death certification was identified, they could immediately query the certifier to modify the death certification. Thus a better quality of cause of death statistics could be attained immediately after the evaluation.
One limitation of this method was that the definition of hospital discharge main diagnosis and the UC do not coincide. Johansson and Westerling further innovatively used ACME (Automatic Classification of Medical Entry) software to solve this limitation, which rendered the monitoring system more robust.11 Many people might not be very familiar with the ACME. In the following, I firstly introduce what ACME is and how it works. Then, I point out some limitations of ACME and the possibility of improvement.
WHAT IS ACME?
To improve the comparability of cause of death statistics among different countries, International Selection Rules for selecting the UC was set by World Health Organisation.12 Nevertheless, the rules leave room for interpretation, which resulted in differences in the selection of UC across countries.13,14 To tackle the problem of inconsistency among coders within and across countries, in the late 1960s and early 1970s the US National Center for Health Statistics (NCHS) developed the ACME computer system to standardise the production mortality statistics.
ACME uses information based on not only the International Classification of Diseases (ICD) codes for each reported condition on the death certificate, but also their actual location on the death certificate. The computer program then applies each international selection rule in sequence to these codes, resulting in a code for a temporary underlying cause (TUC). This TUC code is then subjected to each international modification rule in sequence, finally yielding assignment of a single UC code.15,16 The core of ACME is the Decision Tables, which provide specific relations between one code and another to establish whether the causal sequence is acceptable, highly improbable, or acceptable as a consequence of Rule 3, or whether other modification rules are needed.17 ACME has been used in many countries and broad adoption would certainly improve the comparability of mortality across countries.18 One important feature of ACME is that the logics of selecting the UC for each death certificate could be visualised, which could be used for education and training purposes. The following three examples with different complexity in layouts of diagnoses on death certificates were used to illustrate how ACME processed.
Example 1
I (a) Acute myocardiac infarction (I219)
(b) Hypertension (I10)
(c) Diabetes (E149)
ACME process messages of example 1
01 I219/I10/E149
02 Is I219 due to E149? YES
03 Is I10 due to E149? YES
04 Select Initial TUC = E149—General Principle
05 ACME UC: E149
Example 2
I (a) Congestive heart failure (I509)
(b) Cerebral infarction (I639), endocarditis (I38)
(c) Liver cirrhosis (K746)
(d) Hypertension (I10)
II Chronic obstructive pulmonary disease (J449),
Oral cancer (C069)
ACME process messages of example 2
01 I509/I639 I38/K746/I10*J449 C069
02 Is I509 due to I10? YES
03 Is I639 due to I10? YES
04 Is I38 due to I10? YES
05 Is K746 due to I10? NO
06 No TUC by General Principle—Apply Rule 1
07 Is I509 due to I639? YES
08 Is I639 due to K746? NO
09 Select TUC = I639—Rule 1
10 ACME UC: I639
Example 3
I (a) Congestive heart failure (I509)
(b) Cerebral infarction (I639), endocarditis (I38)
(c) Liver cirrhosis (K746), uremia (N19), Diabetes (E149)
(d) Hypertension (I10)
II Chronic obstructive pulmonary disease (J449), Oral cancer (C069)
ACME process messages of example 3
01 I509/I639 I38/K746 N19 E149/I10*J449 C069
02 Is I509 due to I10? YES
03 Is I639 due to I10? YES
04 Is I38 due to I10? YES
05 Is K746 due to I10? NO
06 No TUC by General Principle—Apply Rule 1
07 Is I509 due to I639? YES
08 Is I639 due to K746? NO
09 Is I639 due to N19? YES
10 Is N19 due to I10? YES
11 Select TUC = I10 - Rule 1
12 Linkage due to position condition I10 I509 I110 MAYBE
13 Linkage with mention of preference I10 I639 YES
14 Linkage due to position preference I10 I38 YES
15 Linkage with mention of combination I10 N19 I120 YES
16 Is I509 due to I639? YES
17 Is I639 due to K746? NO
18 Is I639 due to N19? YES
19 Select TUC = N19—Rule 1
20 Select TUC = I120—Rule C linkage
21 ACME UC: I120
LIMITATIONS OF ACME
Though ACME has been deemed as the de facto international standard for interpreting ICD selection rules, it is not without problems. First limitation was that there were many “MAYBE” causal relations in the decision tables, which needed manual assignments for the UC. Examples were listed as follow:
-
Is K746 (liver cirrhosis) due to A419 (sepsis)? MAYBE
-
Is K746 (liver cirrhosis) due to B169 (hepatitis B infection)? MAYBE
-
Is I698 (sequels of stroke) due to E149 (diabetes)? MAYBE
Is J449 (chronic obstructive pulmonary disease) due to I64 (stroke)? MAYBE
-
Is J189 (pneumonia) direct sequel of I509? MAYBE
-
Is R54 (senility) and I509 (heart failure) combined as R54? MAYBE
If different countries had different decisions for above “MAYBE” cases, this became another source of artefact undermining the comparability of mortality data across countries.
Another limitation, ironically this is in fact the strength of ACME, was the rigid adherence to the selection rules that resulted in the over-coding of mechanism of death (MOD). The MOD is a physiological derangement or a biochemical disturbance produced by a cause of death. Examples include various arrhythmias, renal failure, cardiopulmonary failure, sepsis, and hypovolaemic shock. The cause of death, on the other hand, is a distinct entity, and is aetiologically specific. Examples include cerebrovascular infarction, lung cancer, diabetes mellitus, and alcoholic liver cirrhosis. Because of their lack of aetiological specificity, MOD should not appear on death certificates.19–21 Nevertheless, because medical treatment is often aimed at modifying or ameliorating mechanisms rather than causes, thereby physicians still filled many MODs on death certificate. This poor certification behaviour was fueled by high frequency of incorrect layout of diagnoses on the death certificates. Previous studies revealed that it was very common for physicians to enter two or more diagnoses in the same line in death certificate.22–24 Examples were:
-
I (a) Uraemia, diabetes
-
I (a) Heart failure, liver cancer
-
I (a) Hepatic failure, ischaemic heart disease
Another common certification error was the reverse layout of causal relations. For example, hypovolaemic shock (HS) was due to oesophageal varices bleeding (EVB) and EVB due to liver cirrhosis (LC). A correct layout should put HS in line (a), EVB in line (b), and LC in line (c), nevertheless it was not very uncommon that the certifier might put HS in line (c), EVB in line (b), and LC in line (a). Other examples were:
-
I (a) Acute myocardial infarction
(b) Pneumonia
(c) Sepsis
-
I (a) Stroke
(b) Urinary tract infection
(c) Sepsis
According to international selection rule 2 (for first three examples) and general principle (for last two examples), ACME would select MOD—that is, uraemia, heart failure, hepatic failure, and sepsis as the UC for above examples. Most people will agree that these results were obviously not the original intents of the certifiers. MOD could not provide useful information for prevention.
Luckily many of the above mentioned problems might be solved in Mortality Reference Group (MRG), which was set up by the World Health Organisation with the mandate to issue authoritative instructions on the interpretation of the ICD coding rules and guidelines. The NCHS have pledged themselves to implement the decision of the MRG in ACME decision tables. It is hoped that the modified Decision Tables will be more acceptable to people in most countries.
References
Linked Articles
- In this issue