Cancer statistics terminology explained

Actuarial survival is the estimated survival for a group of people diagnosed with cancer based on cohort data and life tables.

Life tables look at the number of the cohort who are alive and under observation at the beginning of each year, the number dying in each year, the number lost to follow-up each year, the conditional probability of survival for each year, and the cumulative probabilities of survival from the beginning of the observation to the end of each year.

It is most useful when data are only available in grouped categories.

Age-specific rates are generally used to show how incidence/mortality changes with age. These are often reported in five year age groups.

Age-standardised rates are the most common way of reporting cancer statistics. They are generally used to compare populations and overcome the problems caused by different age profiles. They identify real differences between populations which are not because of age.

Cancer attributable risk is the proportion of cancer cases or deaths in a specific population caused by a particular risk factor. This proportion is also known as the population attributable fraction (PAF).

PAFs can be expressed as a percentage, a proportion, or an absolute number of cases or deaths. For example: A study shows that the cancer PAF of ‘risk factor C’ is 0.08. This means that in the population studied, 8% of cancer cases are caused by ’risk factor C’. In this content ‘exposure’ may be defined as any exposure (versus none), or as exposure above/below an optimum level (that level is sometimes defined using Government guidelines).

Theoretically all cancer cases attributable to a risk factor could be prevented by removing exposure to that risk factor. However we acknowledge that it is very difficult to completely remove a risk factor at population level, and so the total number of ‘preventable cancer cases’ based on PAFs is a very ambitious target.

Cohort survival was a commonly used method of calculating and reporting population-based cancer but the disadvantage of this method is that we have to wait a long time for the data to be available and treatments may have moved on so the data is not always reflective of the real situation. Period survival takes the most recent year of follow up and uses the data from the patients diagnosed in different years. It can take into account the information from patients diagnosed more recently than the cohort approach and address the lack of timeliness of cohort survival.

Confidence intervals provide a measure of the reliability of an estimate from the data. The level of confidence is expressed as a percentage, which give the range in which the statistic in question would fall if it were possible to repeat the analysis, for example: upper and lower confidence limits of 95% mean that it is likely to be the same 19 out of 20 times if repeated.

Confidence intervals can also be used to statistically compare figures between two populations (for example, comparing age-standardised rates or survival rates). If the ranges of the two sets of confidence intervals do not overlap, then there is evidence that the two populations are statistically significantly different. However, if the confidence intervals do overlap, then no conclusion can be drawn since the difference between the two populations could be due purely by chance, and formal significance testing would need to be undertaken.

Coverage is the most important measure for screening – it is a measure of what proportion of the eligible population have been screened within whole the screening programme time frame. It should include those screened opportunistically and those that have followed up a screening invitation a long time after the invitation.

Definitions of coverage (e.g. the timeframe used) may vary between countries, so comparison between countries is usually inadvisable.

Crude rates are the simplest method for comparing cases/deaths whilst accounting for population size, but no other aspects, such as age profile.

Cancer is generally more common in the elderly and crude rates are greatly influenced by the proportions of older people in the populations so can mean comparisons between two areas/time periods with different population profiles are misleading. Age-standardised rates are better for comparison.

The number of cancers and tumours identified by screening as a proportion of the population screened.

Expected survival is estimated using data for life expectancy and deaths (life tables).

Various tables exist which makes comparisons difficult.

This can also be called the ‘number of lives saved’ or ‘additional number of people surviving cancer’. It is the difference between a calculation of the number of patients surviving cancer for in one point of time, compared to what would have happened if survival had stayed the same as a previous point in time.

As survival is improving overall, it is usually a positive number.

Cancer incidence is the number of new cases of cancer diagnosed in a specific population within a specific period of time, usually a year. It usually only refers to primary cancers and does not include secondary cancers or recurrences.

Cancer incidence rates are a standard measure of the frequency of cases within a specific period of time relative to a fixed population size, usually 100,000 people or expressed per million if rare.

Cancers diagnosed in the period between routine screening episodes. Interval cancers occur either through failure to detect an abnormality at the time of screening (false negative result), or as a new event after an accurate negative screening result (true interval cancer).

The lead time is the period between when a cancer is found by screening and when it would be clinically detectable. This figure is not directly observable.

The lifetime risk of cancer is an estimate of a person’s chances of being diagnosed with cancer at some point during their life. Lifetime risk calculations use all-cause mortality rates and cancer incidence and mortality rates for a birth cohort followed through their lifetime e.g. people born in 1960 followed as they age.

Lifetime risk estimates are usually expressed as the odds of developing cancer (‘1 in X’), as a percentage, or as the number of people diagnosed per 100 people.

Cancer mortality is the number of deaths from cancer in a specific population within a specific period of time, usually a year. It usually only includes deaths where cancer is mentioned as an underlying cause of death on death certificates.

Cancer mortality rates are a standard measure of the frequency of deaths within a specific period of time relative to a fixed population size, usually 100,000 people or expressed per million if rare.

Net survival estimates the number of people who survive their cancer rather than calculating the number of people diagnosed with cancer who are still alive. In other words, it is the survival of cancer patients after taking into account the background mortality that they would have experienced if they had not had cancer.

Observed survival is the actual percentage of people diagnosed with cancer who are still alive after a specified amount of time. This means it includes death from any cause.

Observed survival is used when the actual numbers of people surviving are required. It is, however, not generally used because a certain proportion of those people who develop cancer are likely to have died anyway within the time period (because cancer patients are commonly elderly) and this underlying background mortality is not taken into account.

An odds ratio (OR) is either the likelihood that people with cancer have been exposed to a particular risk factor, compared with the likelihood that people with cancer have not been exposed to that risk factor (i.e. comparing between cancer patients whether they were or were not exposed) or it is the likelihood that people with cancer have been exposed to a particular risk factor, compared with the likelihood that people without cancer have been exposed to that factor (i.e. comparing between cancer patients and people without cancer who have both been exposed). Which it is depends on whether the research was a retrospective study or a case-control study.

In this context, ‘exposed to’ can also mean ‘born with’ (e.g. a medical condition, or a specific sex), not just exposure to external risk factors.

The results of case-control studies and retrospective studies are often expressed as ORs, because in these study designs researchers ‘work backwards’ from the development of cancer to look for exposure to risk factors in the past. It is important to note that this type of study cannot ascertain causality (i.e. this type of study cannot show that exposure to the risk factor caused cancer to develop), it can only show whether there is an association.

For example: a study shows that the OR of exposure to ‘risk factor B’ in people with cancer is 4.3. This means that people exposed to ‘risk factor B’ have a 330% higher risk of developing cancer, compared with people not exposed to ‘risk factor B’.

Overdiagnosis occurs when a disease is detected via screening but would never have gone on to cause any harm during a person’s lifetime. This can happen with some cancers because they are sometimes slow-growing and unlikely to cause harm.

The variety of methods used to estimate overdiagnosis can produce apparently conflicting results.[1] There are two key methods:

1. The proportion of cancers diagnosed over the lifetime because of screening (which would not otherwise have been diagnosed or caused the patient any harm). This represents the additional demand on the health service caused by screening, and is useful from a public health perspective.
2. The proportion of cancers diagnosed during the screening period because of screening (which would not otherwise have been diagnosed or caused the patient any harm). This represents the probability that a cancer diagnosed is one which would not otherwise have been diagnosed or caused the patient any harm, and is useful from an individual’s perspective.

Overdiagnosis is usually calculated by comparing a screened population with an unscreened population, e.g. across geographical areas (within the same country), between age groups/birth cohorts, or within the same population before and after screening is implemented. However, there are likely to be variations in the underlying true incidence of cancer between these comparison groups, and adjusting for these variations is difficult.[2] Randomised controlled trials (RCTs) of screening were not designed to measure overdiagnosis as an outcome so evidence collected specifically for this purpose is sparse.

The proportion of people with positive screening results who are diagnosed with cancer. Also known as positive predictive value (PPV).

Cancer prevalence provides a 'snapshot' of the number of people who are alive on a particular date, having previously been diagnosed with cancer in a specified geography and within a specified time frame.

Observed cancer prevalence refers to people diagnosed within a specified time frame, while complete cancer prevalence refers to people diagnosed at any time up to the ‘snapshot’ date.

The prevalence figure on this website for all cancers combined is complete prevalence for people diagnosed in the UK up to the end of 2013. The prevalence figures on this website for individual cancer types are observed prevalence for people diagnosed in the UK between 1991 and 2010.

The first time a person has screening within a screening programme is called the prevalent round. Subsequent screening episodes for that person are called incident rounds.

Rates allow comparisons between populations which can’t be done just with the number of cases.

These are often calculated for a year, but sometimes, where the numbers are small and/or there is a lot of fluctuation in the year-on-year values, rates are averaged over several years, often three years.

Relative risk (RR) is the likelihood of cancer developing in people exposed to a particular risk factor, compared with the likelihood of cancer developing in people not exposed to that risk factor. RR can also mean ‘risk ratio’, this is the same as relative risk.

In this context, ‘exposed to’ can also mean ‘born with’ (e.g. a medical condition, or a specific sex), not just exposure to external risk factors. The results of cohort studies and randomised controlled trials (RCTs) are often expressed as RRs, because in these study designs, researchers ‘work forward’ from the risk factor exposure to see if cancer develops in future.

For example: a study shows that the RR of cancer in people exposed to ‘risk factor A’ is 1.95. This means that people exposed to ‘risk factor A’ have a 95% higher risk of developing cancer, compared with people not exposed to ‘risk factor A’.

Relative survival compares the survival of individuals with cancer to those in the general population. Ideally it would be to those without cancer, but this baseline is difficult to obtain. It is similar to the probability of survival from cancer without including any other cause of death.

It was the most commonly used method of calculating and reporting population-based cancer survival but there are various statistical methodologies to calculate it which makes comparisons difficult.

Relative survival can be greater than 100% because it accounts for background mortality. Relative survival greater than 100% indicates that patients in this group have a better chance of surviving one year after diagnosis compared with the general population.

In terms of cancer, risk is the likelihood of developing or dying from cancer, either within a specified period of time or over a lifetime.

The proportion of a screened population which has the disease and tests positive.

For example, a sensitivity of 80% means that for every ten participants with the disease, eight will test positive and the other two will be false negatives. A test with poor sensitivity results in a high proportion of the population with the disease escaping detection. These people will be falsely reassured and could delay presenting important symptoms.

The proportion of a screened population which is disease-free and tests negative.

For example, a specificity of 90% means that nine out of ten people who do not have the disease will have a negative result. One out of ten will have a false positive result and require further assessment before the possibility of disease can be eliminated. A test with poor specificity will have important consequences for the individual, including anxiety and unnecessary follow up.

An ideal screening test would have a high sensitivity (to reduce the number of false negatives) and a high specificity (to reduce the number of false positives). It is usually difficult to achieve this as there is a trade off between the two measures; tightening the criteria for one means in a decrease in the other.

Cancer survival is the percentage of people still alive after a specified amount of time, often 1, 5 or 10 years after a diagnosis of cancer at a specific time (e.g. 2010-11). It usually only refers to primary cancers and does not include secondary cancers or recurrences.

Survival percentages are not ‘rates’ because they are not the number of people in a specified population who survive for the specified period.

Uptake is a performance measure for efficiency of the screening programme. It is a measure of the proportion of people who have been invited that have responded to the invitation. It shouldn’t include those screened opportunistically and those that have followed up a screening invitation a long time after the invitation.

Definitions of uptake (e.g. the timeframe used) may vary between countries, so comparison between countries is usually inadvisable.

For screening programmes where people may be screened following GP referral or self-referral, rather than just following an invitation from the programme (e.g. cervical screening), uptake cannot be calculated because the denominator is unknown.