Flip Flop Flu (part 2)

With the NHS straining under impressive levels of seasonal respiratory infection I decided to take a closer look at influenza and pneumonia

In part 1 of this series I produced a graph of crude mortality for influenza and pneumonia dating back to 1901. We observed the impact of coding changes during a trial period 1984 – 1991 along with implementation of a new system in 2001 that served to de-emphasise the role of respiratory disease by reducing numbers of deaths in official counts. I promised to provide a bit of background to these coding changes so get the kettle on, pop some toast under the grill and pull up a chair…

WHO Rule 3

Officers at the Office for National Statistics tell me that implementation of WHO rule 3 generated these historic coding changes and have kindly pointed me to a ONS national mortality status report that was laid before parliament back in 2006 that usefully summarizes the changes. I have secured a copy of this document, which may be found here, and present a few choice extracts for contemplation:

In general, the main change in introducing automated cause coding was in the interpretation of WHO Rule 3, one of the rules used to select the underlying cause of death. The interpretation of Rule 3 was broadened by OPCS in 1984, so that certain conditions that are often terminal, such as bronchopneumonia (ICD-9 485) or pulmonary embolism (ICD-9 415.1) could be considered a direct consequence of any more specific condition reported. The more specific condition would then be regarded as the underlying cause. This change in interpretation meant that deaths from certain conditions such as pneumonia declined in 1984, while deaths from conditions often mentioned in part II of the death certificate increased. The change in 1993 was thus to move back to the internationally accepted interpretation of Rule 3 operating in England and Wales before 1984. The effects of moving back to this earlier interpretation of Rule 3 have been discussed elsewhere. Deaths assigned to external causes were excluded from the Rule 3 change in 1984 because the procedures for assigning underlying cause of death based on coroners’ verdict were unaffected by WHO rules.

The rule that changes cause of death statistics most is Rule 3. In ICD-10 the list of conditions affected by Rule 3 is more clearly defined than in ICD-9 and is also broader in scope. Its impact is to reduce the number of deaths assigned to certain conditions such as pneumonia and to increase the number of deaths assigned to chronic debilitating diseases. In England and Wales, about 20 per cent of deaths mention pneumonia so the effect of the change in Rule 3 is large.

1984 - OPCS decided to amend its interpretation of WHO Rule 3 in the assignment of underlying cause of death. This amendment is covered in more detail in section 3.2. It resulted in a decrease in the numbers of deaths coded to pneumonia and a few other causes, and an increase in deaths from many other conditions – most of the latter being small increases. The background to this change is given in the annual volume for 1984 in the DH2 series which includes a table assessing the numerical effects of changes, by underlying cause.

1993 - OPCS decided to revert to the internationally accepted interpretation of Rule 3 operating in England and Wales before 1984 (see section 3.2).

2001 - Introduction of the Tenth Revision of the International Statistical Classification of Diseases for coding cause of death on 1 January 2001. This replaced the Ninth Revision used from 1979 to 2000. There are some significant differences between the ICD versions. The main differences are:

• a change in format of the code and an expansion in the number of codes used

• a movement of some diseases and conditions between broad groups called ICD chapters

• changes to the rules governing the selection and coding of the underlying cause of death, especially Rule 3, which has had a large effect (see section 3.2)

..so there you go!

As a consequence, the UK mortality series dating from 2001 is coded under WHO rule 3, which means fewer deaths from influenza and pneumonia than we used to count, which is one reason why the COVID pandemic tends to look like a sore thumb. Authorities, however, have ignored the spirit of this ruling when it comes to counting COVID-19 deaths: this is the died with vs. died of chestnut that took a fair bit of criticism from all quarters before the distinction was made more transparent to the public.

Some may recall that in the early days of data publication COVID deaths were counted according to a rather peculiar 60-day rule rather than the standard 28-day rule; and before this we had an ‘infinite-day’ rule which meant anybody who had fully recovered from COVID but fell under a bus 3 months later would be counted as a COVID death. We should never forget the astonishing level of deceit that was embraced by the ONS.

Another Slice, Vicar?

But this isn’t the worst of it. That phrase ‘automated cause coding’ is another layer of fiddly-widdly complexity that we need to understand, for what has been expressed on the death certificate by an experienced and highly qualified physician may not be what the system churns out at the other end. Below are pertinent paragraphs that I have pilfered from this very handy ONS online document, with the nitty gritty bits marked with bold text:

9. Cause of death coding

Coding the underlying cause of death

Automated cause coding

Since 1993, the majority (approximately 80%) of Office for National Statistics (ONS) mortality data have been coded by automatic cause coding software. Specific text terms from the death certificate are converted to International Classification of Diseases (ICD) codes, and then selection and modification rules (see later in this section) are used to assign the underlying cause of death. Using computer algorithms to apply rules increases the consistency and improves the international and temporal comparability of mortality statistics. The cause coding of deaths certified after inquest is done manually by experienced coders, as the software could not code the free text format used by coroners.

International Classification of Diseases, 10th edition (ICD-10) was introduced in England and Wales in January 2001. Since then, various amendments have been authorised by the World Health Organization (WHO). Amendments may, for example, correct errors in the software supporting automatic coding, accommodate new codes in response to new conditions, such as coronavirus (COVID-19) or incorporate advances in medical knowledge of the relationship between conditions.

Between 2001 and 2010, the ONS used the Mortality Medical Data System (MMDS) ICD-10 version 2001.2 software provided by the United States National Centre for Health Statistics (NCHS) to code cause of death. In January 2011, this was updated to version 2010, which incorporated most of the WHO amendments authorised up to 2009.

The main changes in ICD-10 version 2010 were amendments to the modification tables and selection rules. Overall, the impact of these changes is small, although some cause groups are affected more than others. For further information, see the results of the bridge coding study. There is also another study looking at the impact on stillbirths and neonatal deaths.

On 1 January 2014, we changed the software used to code cause of death to a package called IRIS (version 2013). The development of IRIS was supported by Eurostat, the statistical office of the EU and is now managed by the IRIS Institute hosted by the German Institute of Medical Documentation and Information in Cologne.

IRIS software version 2013 incorporated all official updates to ICD-10 approved by WHO, which were timetabled for implementation before 2014. These updates included changes to the use of codes within the neoplasms (cancer) chapter (ICD-10 codes C00 to D48). In addition, a small number of changes were made to the coding of specific conditions, to bring previous coding practice in line with international coding rules and changes were made to the coding of neonatal deaths and stillbirths.

On 1 January 2022, we again updated the IRIS software to the Multicausal and Unicausal Selection Engine (MUSE) (IRIS version 5.8). The MUSE operates based on internationally agreed decision tables that reflect the most recent version of ICD-10. This system also increases the automation of coding compared with the previous software.

The death certificate (Annex A (PDF, 224KB)) used in England and Wales is compatible with that recommended by WHO. It is set out in two parts. Part I gives the condition or sequence of conditions leading directly to death, while Part II gives details of any associated conditions that contributed to the death but are not part of the causal sequence.

The selection of the underlying cause of death is based on ICD rules and is made from the condition or conditions reported by the certifier, as recorded on the certificate. The underlying cause of death is defined by WHO as the disease or injury that initiated the train of events directly leading to death or the circumstances of the accident or violence that produced the fatal injury.

Selection and modification rules

The selection of the underlying cause of death is generally made from the condition or conditions entered in the lowest completed line of Part I of the Medical Certificate of Cause of Death (MCCD). If the death certificate has not been completed correctly - for example, if there is more than one cause on a single line with no indication of sequence or the conditions entered are not an acceptable causal sequence - it becomes necessary to apply one or more of the selection rules in the ICD-10.

Even where the certificate has been completed properly, there are particular conditions, combinations or circumstances when modification rules have to be applied to select the correct underlying cause of death. On some death certificates, for example, when two or more causes are listed and then linked together, these may point to another cause (not mentioned directly on the certificate) as underlying (an inferred underlying cause). This happens in a minority of cases and these are most commonly related to diseases of the circulatory system and late effects of cerebrovascular disease. In other cases, the underlying cause of death can be selected from Part II of the MCCD.

In summary, the purpose behind the selection and modification rules is to derive the most useful information from the death certificate and to do it uniformly so that data will be comparable between places and times and each death certificate produces one, and only one, underlying cause of death.

14. Checking and validation of registration data at ONS Titchfield

Routine checks in Titchfield

All deaths accepted onto the database that need routine coding are identified and coded as required by the Life Events Processing Branch (LEP). The detailed routine coding falls into five main areas:

• postcoding to give usual residence of deceased

• occupation of the deceased (or spouse, civil partner, or mother and/or father; see Section 2: Information collected at death registration for further details)

• communal establishment coding for place of death of deceased

• place of birth of deceased

• cause of death (see following checks)

Routine automated and manual checks of cause of death data are carried out on all records on a monthly basis. These include:

• checking cause fields against inquest conclusion fields for compatibility

• the presence or absence of original and final cause of death fields

• codes for Office for National Statistics (ONS) cause groups are present for neonatal deaths and absent for non-neonatals (see Section 11: Childhood mortality; subsection: Neonatal deaths)

• validity of suicides at very young ages

• mentioned conditions on death certificate are compatible with sex

• the derived underlying cause of death is mentioned in Part I or Part II of the death certificate

Checks before and after extraction of data for analysis

The first of these are carried out as a final check of what is held on the deaths database before an annual extract of data is taken. These comprise frequency checks for a range of fields, covering age, sex, underlying cause and area of residence. Also checked are possibly incorrect combinations of fields. Any apparent errors or inconsistencies result in checks of individual cases by coders who make amendments, as required. Some of these checks are also carried out routinely every month.

Brain Ache!

I appreciate that digesting all of that may have induced brain ache, so here’s a slice of Somerset apple cake:

What I’d like folk to appreciate is that there is a chasm between what is actually written on death certificates and what numbers may be published by the ONS. That chasm is bridged by some rather fancy software that relies on AI to make decisions on causality to produce one and only one conforming cause of death regardless of what the certifying physician may think and regardless of the clinical reality of multi-causal death.

When it comes to automated coding and the pandemic I doubt very much whether we’ve seen anything approaching the truth, and we’re not likely to ever see truthful ONS figures for vaccine harm even under FOI, since all figures, tables and charts are subject to a great hive mind buzzing and brimming with AI that is (US) built to specification to ensure the narrow-minded narrative dictated by WHO will always play out. Flip flop flu indeed!

Kettle On!