HES data are generated from the hospital records that flow from the providers of hospital care (the NHS hospital trusts) to the commissioners of care. There are over 12 million records of admitted patient care each year, so there are likely to be some errors in these records. The HES system cleans the data prior to their publication. This involves ensuring that all the codes and dates are valid (i.e. they are in the correct format and in the correct range) and overwriting invalid codes and dates with "unknown" indicators. HES also checks for anomalies in the data, for example where the record shows a male who has given birth, and either corrects the data if there is sufficient evidence that a code can be corrected, or overwrites the code as invalid if there is insufficient evidence to determine what has caused the anomaly. Furthermore, some codes are inherently less reliable than others, for example the ethnic codes are not always completed, and where they are, the patient selects the code. HES publishes for each trust that show where and to what extent HES has detected data deficiencies. Where HES detects data that may be corrected manually, for example where the trust has consistently used an incorrect provider code, this may be done provided the trust agrees.

HES can provide data on an annual basis (1 April to 31 March) from 1989-1990 to the present. However, making meaningful comparisons over time requires knowledge of the many changes that have occurred in both the organisation of the NHS, and the methods used to record information about admitted patient care.

Here are just two of the factors that may need to be taken into account:

• health authorities ceased to exist in April 2002 and the healthcare commissioning role passed to primary care trusts (PCTs);
• Diagnoses were coded using the 9th revision of the International Classification of Diseases (ICD9) up until April 1995, but from this date the 10th revision, (ICD10) has been used. ICD10 is a complete re-working of the classification and has an entirely different code structure.

HES includes data that show whether a patient died in hospital. However, the cause of death is not recorded and may be unrelated to the patient's diagnosis or treatment. Therefore, we recommend that analysts do not use HES death data for drawing any conclusions about treatment. The Office for National Statistics (ONS) collects cause of death as part of the deaths registration process and The Information Centre for health and social care is working with the National Centre for Health Outcomes Development (NCHOD) to develop a linked file that will associate deaths with the information held in HES. Initially, this will only include information on the date of death.

Yes: the International Classification of Diseases (ICD) provides special diagnosis codes that are used to give additional information about the cause of the patient's condition. For example, if the primary diagnosis indicates a broken arm, there may also be another code saying that this was due to a fall from a tree. HES records all of these codes and they can be used to help analyse the reasons why people have required hospital treatment. However, it is only necessary to record information on the cause of a patients condition the first time they are treated, so any subsequent admissions or episodes relating to the accident may not include information on the cause.