Mobile Phone Use for Contacting Emergency Services
Mobile Phone Use for Contacting Emergency Services
We constructed a population cohort by record linkage. Mortality (at the scene, in the emergency department [ED], and during hospitalization), transfer to the ED, admission (to inpatient care and to the intensive care unit [ICU]), and corresponding length of stay (inpatient care and the ICU) were analyzed by initial exposure (mobile phone vs. landline), controlling for potential confounders described below in "Outcome Measures."
Oxfordshire Ambulance Service National Health Service (NHS) Trust (OAS) was responsible for providing all ambulance dispatches in emergencies in the UK county of Oxfordshire. In 2004–2005, Oxfordshire had an area of 1006 square miles, with a resident population of 608,000, an estimated transient tourist population of more than 9.3 million annually, with a significant rural component. Each emergency call received by the OAS is logged from the point of call to the point of delivery of the patient to the ED through an electronic system. The OAS dataset covered the period from January 1995 to June 2006 and included the telephone number from which the call originated, the address location of the caller, characteristics of the patient and the nature of the patient's condition, and the activity of the attending ambulance crew. Emergencies were coded and then assigned a series of alphanumeric codes based on the UK Department of Health-approved Advance Medical Priority Dispatch System, which uses structured protocols and systematic questioning of the 999 (emergency phone line) caller, and is used by over 75% of ambulance services. Only those dispatches designated as Code Red (i.e., immediately life-threatening) were included in this study. Postcodes were designated to the incident scenes from May 2001 and Ordnance Survey (OS) coordinates from April 2001. Before that, a local area coding system had been used. The OAS dataset covers the period from January 1995 to June 2006.
The Oxford Radcliffe Hospital NHS Trust (ORH) serves the Oxfordshire population through EDs at the John Radcliffe Hospital (JRH) in Oxford and the Horton General Hospital (HH) in Banbury, in the north of the county. Since October 2000 there have been, in addition, direct emergency admissions to a Medical Assessment Unit, and to a Surgical Emergency Unit at the JRH from November 2002. Data were available from the Patient Administration Systems (PAS) for each ED and separately for inpatients. For JRH, PAS data were available starting January 1995, but the HH ED PAS data were only available since March 1999.
In the absence of a unique identifier linking ambulance and hospital records, an algorithm was developed to identify each OAS call, the most likely resulting ED arrival (if any), and the most likely subsequent inpatient admission (if any). The availability and quality of data in several fields of the datasets varied greatly over time. For example, patient names were not recorded in the OAS database after July 2001, and we deduced that many recorded before that date were the caller's rather than the patient's name. A set of heuristics was developed for matching the following OAS fields against corresponding fields in the ORH databases: patient's first and last name, sex, age, and presenting complaint; the location to which the ambulance was dispatched; the hospital to which the patient was delivered; and the time of arrival at the hospital.
The data about patients in the OAS database were entered during an emergency phone conversation, often with a panicked caller. The result is that much of the free-text data, such as proper names, are phonetic transcriptions of what the caller is saying. We used a variation of the Soundex algorithm (www.sound-ex.com) to help with matching such fields.
Each of the above criteria was allocated a weight according to its usefulness in identifying a match. The set of weights was developed during a manual matching process of 100 randomly selected records. For each field, the weights of matching fields were summed and the weights of positively non-matching fields were subtracted from the sum. The result was a matching score for each pairing of an OAS record and at least one ORH record. At this stage, there were several cases of an OAS record matching several ORH records, and vice versa. This was resolved by selecting the most likely match as any match whose score was significantly higher than all the others. The result was a pairing of the hospital-delivered OAS records with a uniquely corresponding ORH ED or inpatient record. We randomly inspected a small number of records to assess the reliability of the matching at each stage of development of the algorithm. Unmatched records, and records with a very low matching score, were discarded. Additional information on the matching algorithm is available as a web appendix (Appendix 1).
The outcome measures of the study were mortality, transfer to ED, admissions to the hospital, and length of stay. Mortality was measured in terms of deaths at the scene of the incident (deaths before arrival and during intervention of ambulance service), at discharge from the ED, and at discharge from the hospital; transfers from the scene to the ED were reported in ED records; admissions were measured in terms of inpatient admissions and admission to the ICU; and lengths of hospital stay relating to inpatient admissions and intensive care unit were also recorded.
Mobile phone exposure was determined by the recorded telephone number of the call made to emergency services. Mobile phone telephone numbers were identified by the prefixes "03," "04," "05," "06," "08," and "09" before September 30, 1999, and the prefix "07" in all years, excluding all known landline numbers in the area with the same prefixes.
To account for potential confounders, the following variables were included in the analysis:
We carried out a series of multivariate analyses using logistic regression to determine the association between the reporting of emergencies using mobile phones, and all deaths and admission outcomes, while adjusting for potential confounders. To account for the skewed nature of the data, generalized linear models assuming a gamma distribution for the data with a canonical link function (log) were used to estimate the mean difference in length of stay, in overall inpatient care, and in ICU between calls originating from mobile phones and landlines, while adjusting for potential confounders. The probabilities relating to individual outcomes were estimated from the regression model. All analyses were carried out in Stata 10.0 (StataCorp LP, College Station, TX).
Materials and Methods
Study Design
We constructed a population cohort by record linkage. Mortality (at the scene, in the emergency department [ED], and during hospitalization), transfer to the ED, admission (to inpatient care and to the intensive care unit [ICU]), and corresponding length of stay (inpatient care and the ICU) were analyzed by initial exposure (mobile phone vs. landline), controlling for potential confounders described below in "Outcome Measures."
Setting
Oxfordshire Ambulance Service National Health Service (NHS) Trust (OAS) was responsible for providing all ambulance dispatches in emergencies in the UK county of Oxfordshire. In 2004–2005, Oxfordshire had an area of 1006 square miles, with a resident population of 608,000, an estimated transient tourist population of more than 9.3 million annually, with a significant rural component. Each emergency call received by the OAS is logged from the point of call to the point of delivery of the patient to the ED through an electronic system. The OAS dataset covered the period from January 1995 to June 2006 and included the telephone number from which the call originated, the address location of the caller, characteristics of the patient and the nature of the patient's condition, and the activity of the attending ambulance crew. Emergencies were coded and then assigned a series of alphanumeric codes based on the UK Department of Health-approved Advance Medical Priority Dispatch System, which uses structured protocols and systematic questioning of the 999 (emergency phone line) caller, and is used by over 75% of ambulance services. Only those dispatches designated as Code Red (i.e., immediately life-threatening) were included in this study. Postcodes were designated to the incident scenes from May 2001 and Ordnance Survey (OS) coordinates from April 2001. Before that, a local area coding system had been used. The OAS dataset covers the period from January 1995 to June 2006.
The Oxford Radcliffe Hospital NHS Trust (ORH) serves the Oxfordshire population through EDs at the John Radcliffe Hospital (JRH) in Oxford and the Horton General Hospital (HH) in Banbury, in the north of the county. Since October 2000 there have been, in addition, direct emergency admissions to a Medical Assessment Unit, and to a Surgical Emergency Unit at the JRH from November 2002. Data were available from the Patient Administration Systems (PAS) for each ED and separately for inpatients. For JRH, PAS data were available starting January 1995, but the HH ED PAS data were only available since March 1999.
Data Collection and Processing
In the absence of a unique identifier linking ambulance and hospital records, an algorithm was developed to identify each OAS call, the most likely resulting ED arrival (if any), and the most likely subsequent inpatient admission (if any). The availability and quality of data in several fields of the datasets varied greatly over time. For example, patient names were not recorded in the OAS database after July 2001, and we deduced that many recorded before that date were the caller's rather than the patient's name. A set of heuristics was developed for matching the following OAS fields against corresponding fields in the ORH databases: patient's first and last name, sex, age, and presenting complaint; the location to which the ambulance was dispatched; the hospital to which the patient was delivered; and the time of arrival at the hospital.
The data about patients in the OAS database were entered during an emergency phone conversation, often with a panicked caller. The result is that much of the free-text data, such as proper names, are phonetic transcriptions of what the caller is saying. We used a variation of the Soundex algorithm (www.sound-ex.com) to help with matching such fields.
Each of the above criteria was allocated a weight according to its usefulness in identifying a match. The set of weights was developed during a manual matching process of 100 randomly selected records. For each field, the weights of matching fields were summed and the weights of positively non-matching fields were subtracted from the sum. The result was a matching score for each pairing of an OAS record and at least one ORH record. At this stage, there were several cases of an OAS record matching several ORH records, and vice versa. This was resolved by selecting the most likely match as any match whose score was significantly higher than all the others. The result was a pairing of the hospital-delivered OAS records with a uniquely corresponding ORH ED or inpatient record. We randomly inspected a small number of records to assess the reliability of the matching at each stage of development of the algorithm. Unmatched records, and records with a very low matching score, were discarded. Additional information on the matching algorithm is available as a web appendix (Appendix 1).
Outcome Measures
The outcome measures of the study were mortality, transfer to ED, admissions to the hospital, and length of stay. Mortality was measured in terms of deaths at the scene of the incident (deaths before arrival and during intervention of ambulance service), at discharge from the ED, and at discharge from the hospital; transfers from the scene to the ED were reported in ED records; admissions were measured in terms of inpatient admissions and admission to the ICU; and lengths of hospital stay relating to inpatient admissions and intensive care unit were also recorded.
Mobile phone exposure was determined by the recorded telephone number of the call made to emergency services. Mobile phone telephone numbers were identified by the prefixes "03," "04," "05," "06," "08," and "09" before September 30, 1999, and the prefix "07" in all years, excluding all known landline numbers in the area with the same prefixes.
To account for potential confounders, the following variables were included in the analysis:
Age and sex of the patient
Year of incident
Type of incident – medical or injury
Urban-rural indicator – the geographical location of the incident was dichotomized using postcodes or OS grid reference data
Time of day the call was made – categorized as daytime (8:00 a.m. to 5:00 p.m.), evening (5:00 p.m. to midnight), and overnight (midnight to 8:00 a.m.)
Ambulance transport time (minutes) – time of arrival at the hospital minus time of departure from scene of incident
Ambulance transport distance – calculated as distance between the postcodes or OS grid reference for the location of incident and the hospital, categorized into 0–10 Km, 11–20 Km, and > 20 Km.
Primary Data Analysis
We carried out a series of multivariate analyses using logistic regression to determine the association between the reporting of emergencies using mobile phones, and all deaths and admission outcomes, while adjusting for potential confounders. To account for the skewed nature of the data, generalized linear models assuming a gamma distribution for the data with a canonical link function (log) were used to estimate the mean difference in length of stay, in overall inpatient care, and in ICU between calls originating from mobile phones and landlines, while adjusting for potential confounders. The probabilities relating to individual outcomes were estimated from the regression model. All analyses were carried out in Stata 10.0 (StataCorp LP, College Station, TX).
Source...