Agnieszka Fihel writes about the quality of multi-morbidity data and its importance in mortality research, based on her study of Poland.

What is multi-morbidity?

The notion of multi-morbidity became very popular in the time of the COVID-19 pandemic. But what does multi-morbidity, in other words co-morbidity, actually mean? The term is usually used to describe the coexistence of several chronic diseases that may have similar determinants or that result from the physiological process of individual ageing (Anderson, 2011). An example of someone experiencing multi-morbidity is an obese person addicted to smoking, who suffers from diabetes, hypertension, and chronic obstructive pulmonary disease.

Many health and demography experts recognize multi-morbidity as an important and current topic. In ageing populations, a single cause – regardless of how precisely determined – no longer adequately describes the morbidities responsible for a large proportion of deaths (Dorn and Moriyama, 1964: 401). Pinpointing only one cause of death means that a large part of the pathological condition is not considered in the analysis (Israel et al., 1986). The importance of multi-morbidity has grown due to the 2020 pandemic, when it became clear that COVID-19 is most dangerous to those already suffering from multi-morbidities (Mair et al., 2020).

Unfortunately, there are often serious obstacles to investigating the topic at hand. The reason is that only a few countries make data on multi-morbidity from national statistical systems available to researchers and the larger public. This limits the spectrum of analysis one can perform. For example, in demographic studies, the combination of different morbidity conditions is investigated only in relation to mortality based on information derived from the death certificates – so-called multiple causes of death (MCoD) studies.

These documents include the list of the diseases and medical conditions that influenced the health status of a deceased person, i.e. those that contributed to worsening health and (potentially) led to death. The main condition behind death is called ‘the underlying cause of death’, whereas all other conditions that deteriorated health status are called ‘contributing causes’. For instance, pancreatic cancer may lead to emerging liver metastasis, diabetes due to dysfunctionality of the pancreas and, in the final hours, anemia. In this case, pancreatic cancer initiated the chain of conditions leading to death and constituted the underlying cause of death, while all the other conditions were contributing causes.

The rise of research on multiple causes of death (MCoD)

Though the first MCoD analyses were published several decades ago, the approach gained popularity only recently. This was possible mostly thanks to technological improvements implemented in national statistical systems in regard to cause-of-death data collection.

Consequently, as accessibility of cause-of-death data improved, novel methods and measures of analysis were proposed (Désesquelles et al., 2010; Egidi et al., 2018) and scientific networks (e.g. the Multiple Causes-of-Death Network) were established.

However, what is actually the goal of MCoD research? Put simply, it is to identify the links between chronic diseases and pathologies. The main objectives are especially:

• the investigation of the whole morbidity process that leads to death, and
• highlighting the importance of diseases and risk factors that are rarely registered as the underlying causes, but are frequently certified as contributing causes (e.g. asthma, diabetes or hepatitis).

The curious case of Poland

So far detailed MCoD studies have been performed for such countries as France, Italy, the United Kingdom and the United States. My recent research, to the best of my knowledge, is the first analysis of this kind conducted in Poland.

Why is it so?

The main reason is the relatively low-quality of Polish cause-of-death data due to the frequent assignment of deaths to unknown and ill-defined causes. To make things worse, these problematic causes represent only part of a larger category of so-called ‘garbage’ codes (GCs), defined as causes of death that are not useful in analyses of public health and mortality.

The quality of Polish data is unsatisfactory despite the World Health Organisation’s (WHO) recommendations and actions implemented by public institutions in order to improve the quality of cause-of-death data in Poland. The share of deaths assigned to unknown and ill-defined causes has remained stable since 2000 and exceeds 6%. At the same time the share of all GCs exceeds 20% and the picture is even worse for older groups (Figure 1). Consequently, the quality of cause-of-death data is unequivocally assessed as low, and Poland is regularly excluded from international mortality analyses performed by the WHO.

Figure 1. Proportion of deaths due to all Garbage Codes and due to selected Garbage Codes as underlying causes by age, Poland 2013

2013 – the year with exceptionally exhaustive data

As in many other countries, the information on multi-morbidity conditions in Poland are destroyed once the underlying cause of death is registered and validated. However, the 2013 data were an exception. For the purpose of testing the automatic coding system of underlying causes of death, the scans of death certificates in 2013 were preserved and later made available for this research. Thus, information of all contributing causes accompanying the underlying causes could be analysed at the individual level. This study uses data from:

• 387,988 permanent Polish residents, deceased within Polish territory in 2013;
• only 10% death certificates had no additional MCoD mentioned.

Among 90% death certificates with at least one MCoD, the data indicated that: 11% reported one MCoD, 61% reported two, 18% reported 3 or more MCoD.

Don’t count your chickens until they’ve hatched

The 2013 data offer a unique opportunity to research the Polish context. Unfortunately, the informative value of these data was relatively low as many of them concerned ill-defined, unknown or other garbage causes. For this reason I performed data-quality analysis. Logistic regression demonstrated that the probability of each additional contributing mention that was not a garbage code was higher when:

• death occurred in the hospital, or
• the underlying cause was well-defined (non-garbage) (Table 1).

Conversely, when death took place in home, the probability of an additional well-defined mention was almost two times lower than in the hospital. Also, in the event that cardiovascular disease was the underlying cause of death, the chance of each additional well-defined mention was higher than for neoplasms and other (excluding external) causes.

Table 1. Ordered logistic regression results (Odds Ratios) for the number of non-garbage contributing mentions^a,b,c, Poland 2013

Variable	Number of non-garbage contributing mentions
Sex (ref. men)	1.073***
Age	0.999***
Place of death (ref. hospital)    Other medical institution    Home    Other	0.893*** 0.530*** 0.540***
Certifying person (ref. doctor based on post-mortem autopsy)   Doctor without autopsy   Other medical personnel	0.879*** 0.798***
Underlying cause (ref. non-garbage code)    Garbage Code	0.336***
Underlying cause (ref. cardiovascular)    Neoplasms    External causes    Other	0.631*** 1.389*** 0.843***
Pseudo R2	0.0509
N	387,988

Rare diseases mean high-quality data

Interestingly, the average number of non-garbage contributing codes was highest when the underlying cause was a condition originating in:

• the perinatal period,
• congenital malformations,
• deformations, or
• chromosomal abnormalities (Figure 2)

The above list includes relatively rare cases. Most probably, these deceased patients had detailed medical documentation and had been treated in highly specialised institutions, which favour proper diagnosis and description of well-defined diseases. In turn, the most frequent underlying causes of death – like neoplasms and diseases of respiratory or circulatory systems – were rarely registered with non-garbage contributing mentions. On average, out of two death certificates with a respiratory or cardiovascular disease as the underlying cause, only one included a non-garbage contributing code.

Figure 2. The average number of garbage and of non-garbage contributing mentions by underlying cause group ^a, Poland 2013

Underlying versus contributing death causes

As already mentioned, the usual objective of MCoD analysis is to demonstrate how the diseases rarely registered as underlying causes contribute to the morbidity process leading to death. To this end, Désesquelles et al. (2010) proposed to calculate the age-standardised mortality rates for:

• conditions registered as underlying cause of death, and
• the same conditions mentioned in death certificates as underlying or contributing causes of death.

The ratio of the latter to the former is the so-called Standardized Ratio of Multiple to Underlying cause (SRMU). Whenever a disease is frequently selected as the underlying cause but not as the contributing cause, the SRMU is low.  A SRMU equal to 1 represents a situation when a disease is selected exclusively as the underlying cause; for instance, HIV is almost always certified as the underlying, not contributing cause of death. The SRMU equals 2 when a disease is selected as often as the underlying and the contributing cause. In turn, when a disease is rarely assigned as the underlying cause of death, but often assigned as the contributing cause of death, which usually refers to diabetes, the SRMU is higher than 2.

As shown in Figure 2, the SRMUs in Poland turned out to be the highest for the following ICD-10 chapters:

• diseases of the blood and blood-forming organs – which were indicated almost 17 times more often as the underlying or contributing causes than as the underlying causes only,
• diseases of the skin and subcutaneous tissue (almost 15 times more),
• diseases of the genitourinary system (5 times more),
• mental and behavioural disorders (4.5 times more),
• infectious and parasitic diseases (4 times more).

Within the more detailed groups of causes, the SRMUs are the most elevated for dementias (excluding Alzheimer’s; SRMU of 20), hyperplasia of prostate (19), septicaemia (10), renal failure (9), obesity and diseases of the thyroid gland (both almost 8). For these diseases and groups of causes, loss of information is greatest when the single-cause-of-death approach is applied. These results are consistent with the MCoD studies obtained for other countries.

Contrary to expectations, the SRMUs are relatively low for some diseases typical of ageing societies, such as neurological disorders (Parkinson’s and Alzheimer’s disease), acute and chronic respiratory diseases, other infectious and parasitic diseases, senility, and some chronic diseases, such as cerebrovascular diseases and diabetes mellitus. At the same time, some deaths due to these diseases are hidden in the ‘other’ categories (mainly unknown or unspecified) of the ICD-10 chapters. Therefore, in order to investigate diseases typical of contemporary populations, it is necessary not only to focus on typical diseases of old age, but also to identify other specific causes, such as volume depletion or vascular dementia that are frequent morbidity conditions found in older adults.

Allowing the accurate recording of multi-morbidity conditions advances our understanding of the interrelations between different diseases. This body of research provides grounds for better, more effective, health policies dedicated to patients with chronic diseases. For instance, the MCoD data aggregated at the national level indicate the extent to which diabetes patients suffer from hypertension and how deadly the coexistence of these two conditions can be. This constitutes indisputable recommendations for stakeholders to improve treatment dedicated to those suffering from these conditions.

The post is based on the paper Investigating multiple-cause mortality in Poland by Agnieszka Fihel. The paper was published in Studia Demograficzne, Issue 2(178), 2020.

About the author: