- Email: [email protected]
Population-based prediction of trauma volumes at a Level 1 trauma centre
Injury, Int. J. Care Injured (2004) 35, 1239—1247
Population-based prediction trauma volumes at a Level 1 trauma centre Greg J. Beilmana,b,*, Jodie H. Taylorb,c, Lisa Joba, Jesse Moend, Aaron Gullicksone a
Department of Surgery, North Trauma Institute, North Memorial Medical Center, Robbinsdale, MN, USA Department of Surgery, Division of Surgical Critical Care, University of Minnesota, MMC 11, 420 Delaware Street SE, Minneapolis, MN 55455, USA c Department of Surgery, Hennepin County Medical Center, Minneapolis, MN, USA d Arithmancer Demographic Consulting, Minneapolis, MN, USA e Department of Demography, University of California, Berkeley, CA, USA b
Accepted 23 March 2004
KEYWORDS Demographics; Elderly; Stochastic; Injury
Summary Objective: With an ageing US population, the demographics of traumatic injuries are being significantly altered. Census projections predict that the number of Americans over age 65 will double in the next 20 years. We used stochastic methods to forecast trauma admissions in order to predict the effects of such demographic changes at our trauma centre. Methods: Age- and sex-related rates of traumatic admission were determined using population statistics and trauma registry data from 1994 to 1999. These rates were then projected from 2000 to 2025 based on both the Lee—Carter and random walk with drift methods. Stochastic population projections were made and paired with the projected trauma rates, allowing estimation of total trauma volume. Results: Trauma rates were predicted to increase for most age groups. Trauma admissions are predicted to increase 57% by 2024. By 2019, 50% of trauma admissions will be 60 or older. Conclusions: Our trauma volume is expected to increase 57% by 2024, an increase of 2% per year. More of this volume will consist of elderly patients, potentially requiring increased health-care resources. ß 2004 Elsevier Ltd. All rights reserved.
Introduction Trauma is traditionally a disease of the young. In patients who are less than 45 years of age, trauma *Corresponding author. Tel.: þ1-612-625-7911; fax: þ1-612-626-0439. E-mail address: [email protected] (G.J. Beilman).
remains the leading cause of death in the United States. Although trauma is only the seventh leading cause of death in patients 65 and older, the death rate from trauma actually increases with age. In these older patients, the rate of injury-related death is 103 per 100,000 population, and for patients aged 75 and older, the injury death rate climbs to 161 per 100,000 population, three times higher than the rate
0020–1383/$ — see front matter ß 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.injury.2004.03.018
1240 for all ages combined.10 Furthermore, time spent in the hospital for similar injuries is two times greater in patients older than 65 than those under 45, mainly because of the increased incidence of complications.10,14,15 This drives an enormous increase in resource utilization for these older patients. The ageing of the US population due to the ‘‘baby-boom’’ generation may soon place a great strain on the capacity of some trauma centres to provide adequate and reliable trauma care. It has been shown that elderly patients over age 70 are almost five times more likely to die from traumatic injury than younger patients even when corrected for severity of injury.3 These deaths typically occur later after injury.3 This is believed to be related to a decline in immunologic and physiologic function and an increased prevalence of pre-existing disease.3,17 Furthermore, in Minnesota, approximately 20% of the elderly population lives in rural areas where there is no access to a Level 1 trauma centre.12 Rural patients who are severely injured require transfer to one of the three Level 1 trauma centres in the neighbouring cities of Minneapolis and St. Paul. It has been shown in other studies that rural patients are more likely to die from traumatic injury than their urban counterparts.1,2,17,21 As the population of Minnesota ages, the number of elderly patients in rural areas is likely to increase, even when accounting for migration. If there is a potential for rural geriatric patients to do poorly when compared to urban geriatric patients, then there may soon be a need for increased availability of trauma care to these patients. There are statistical techniques available to predict future population changes. A stochastic technique has been applied to projections in larger populations such as the United States.7—9 We used this technique along with trauma admission data from the past 5 years to forecast future changes in trauma volume at our Level 1 trauma centre. This project was devised in order to assist in planning for trauma care in the next decades.
Materials and methods Prediction of population trajectories The cohort component method, a common method in demography, was used to project the overall population and its age distribution. This method uses estimates of future fertility, mortality, and migration rates to project birth-cohorts forward in time and to replace them as new birth-cohorts enter the population. In order to incorporate the uncertainty of these future vital rates, a stochastic
G.J. Beilman et al.
method was used. Using this modelling technique, different rates of such variables are sampled from a probability distribution, and a large number of possible population trajectories are created. Stochastic forecasts predict rates which fluctuate realistically, and therefore do not artificially force populations toward a stable configuration.9 This is particularly important when the age distribution of the population matters, as it does in this case. Further, stochastic methods allow values such as the median and 95% confidence interval to be calculated. This model was applied to forecast both Hennepin County (i.e. Minneapolis) and the rest of Minnesota separately in order to predict both local and regional trauma volume, as this hospital serves as a site for both metropolitan and state-wide acute care.
Estimation of population Based on the 2000 US Census, the current estimated population of Hennepin County (i.e. urban Minneapolis) is 1.1 million, and the state of Minnesota (including suburban Minneapolis and neighbouring St. Paul) approaches five million total citizens.12 To facilitate estimation of future population size, the components of population were broken into fertility, mortality, and migration. Age-specific fertility and mortality rates were estimated based on birth counts from 1980 to 1998 and death counts from 1970 to 1999, both available from the Minnesota Department of Health.12 United States Census Bureau estimates of mid-year population size by age from 1970 to 1999 were used as the denominator for these rates.12 However, the Census Bureau underestimated growth in Hennepin County during the 1990s requiring correction of these population counts from 1990 to 1999 based on the observed counts of births and deaths by age each year and net migration implied by the 2000 Census population counts. Data on migration variability was gathered from the Internal Revenue Service.5 An age-specific pattern of net migration for both Hennepin County and the rest of Minnesota was developed from the Public Use Microdata Sample of the 1990 US Census.19 A model described by Lee and Carter was used to forecast each component within reasonable bounds.6,7 In order to predict possible trajectories of population changes, it was assumed that the age-specific patterns of fertility and migration would remain the same. Using the stochastic method, 1000 trajectories of fertility, mortality, and migration were generated, and these trajectories were then used to create 1000 population projections. Each of
Population-based prediction of trauma volumes
the forecasts was for 25 years starting from the 1 January 2000 population.
1241
Results Components of the forecasts
Estimation of trauma volume Retrospective data was collected on all trauma patients from our North Memorial Medical Centre trauma registry over a 5-year period from 1 January 1994 to 1 January 1999. Trauma rates of admission to our trauma centre were developed by using US Census Bureau county estimates as our exposure base, corrected for the 1990s.12 The data was further separated into trauma rates by sex, age, and region. Then a random walk with drift method was used to predict future trauma volume. In order to facilitate this forecast, we made two assumptions: (1) trauma rates will not climb higher than 125% of the highest rate or drop lower than 80% of the lowest rate of rates noted between the years 1994 and 1999; (2) percent of trauma seen at each hospital site will not change. The population forecasts created were then applied to the trauma volume forecast to estimate changes in trauma volume.
Fertility rate The historical and projected total fertility rate for both Hennepin County (Minneapolis) and the rest of Minnesota data are graphed in Figure 1a and b. Hennepin County’s fertility rate shows a definite upward trend in the 1980s that is not apparent for the rest of Minnesota. Since forecasts using this trend would have resulted in an exceedingly high fertility rate in the long-term, it was decided to project ‘‘trendless’’ fluctuation in this case. The median projections for both Hennepin County and the rest of Minnesota are based on 1000 iterations and predict that the total fertility rate will average out at approximately two births per woman over her reproductive life-cycle in both regions, with an upper and lower 95% confidence interval of 1.6 and 2.3 in 2024. These confidence intervals are somewhat smaller than those for the United States as a whole, due to the shorter time series used in this research.
Figure 1 (a) Historic and projected total fertility rate (TFR), Hennepin County, Minnesota. Median projected TFR for 2024 ¼ 2:0 births per woman. Upper and lower 95% confidence intervals ¼ 1:6 and 2.3, respectively. (b) Historic and projected total fertility rate (TFR), rest of Minnesota (excluding Hennepin County). Median projected TFR for 2024 ¼ 1:9 births per woman. Upper and lower 95% confidence intervals ¼ 1:6 and 2.3, respectively.
1242
G.J. Beilman et al.
Figure 2 (a) Female life expectancy, Hennepin County, Minnesota. Median age in 1999 ¼ 82:6 years. Median projected age in 2024 ¼ 87:3 years. (b) Male life expectancy, Hennepin County, Minnesota. Median age in 1999 ¼ 77:6 years. Median projected age in 2024 ¼ 83:2 years. (c) Female life expectancy, rest of Minnesota. Median age in 1999 ¼ 81:5 years. Median projected age in 2024 ¼ 85:0 years. (d) Male life expectancy, rest of Minnesota. Median age in 1999 ¼ 76:4 years. Median projected age in 2024 ¼ 80:9 years.
Life expectancy The Lee—Carter method was used to model the historical mortality time series and a random walk with drift method was used to project mortality
rates into the future.7 The life expectancy forecasts for the different regions were forced to co-vary perfectly in each year such that the differences observed in 1999 between regions and sexes were
Population-based prediction of trauma volumes
1243
maintained. This constraint was added due to the high covariation of the historical time series. The historical and projected life expectancy for Hennepin County females and males is shown in Figure 2a and b, respectively. A median projection based on 1000 iterations is shown which forecasts the anticipated life expectancy for females and males to 2024. There is a definite trend upward in life expectancy for both Hennepin County females and males which continues in the forecast. Life expectancy data for females and males in the rest of Minnesota is shown in Figure 2c and d, respectively. As with the Hennepin County data, the median projection for the rest of Minnesota predicts that both the female and male populations will live longer. Migration Age-specific rates of net migration were estimated from the 1990 Census Public Use Microdata Sample for Minnesota.19 At approximately age 20—24 years, there is a net positive migration for both females and males into Hennepin County (Figure 3a). (This may represent an influx of post-college graduates
entering the urban workforce.) After this large flux, the net migration rate declines and then plateaus in the slight negative range. In the late teen years, the rest of Minnesota sees an increase in outward migration as children leave home (Figure 3b). A large upswing is then evident in the late twenties to thirty age group for both females and males. The net migration for the rest of Minnesota then remains slightly positive until another increase occurs in the elderly population (1970s—1980s). The Internal Revenue Service provides information based on tax returns on total in and out migration at the county level for every county in the United States.5 Combining these data with the age schedules of net migration rates and population estimates produced estimates of the ‘magnitude’ of net migration in a given year. These ‘magnitudes’ were used to estimate future variability in net migration by simple random drawing.
Final population projection Taking all components into consideration, the populations of Hennepin County and the rest of Minnesota
Net Migration Rate, Hennepin County, 1990 (Combined Sexes) 0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 -0.05 0
20
40
(a)
60
80
100
Age
Net Migration Rate, Rest of Minnesota, 1990 (Combined Sexes) 0.015 0.01 0.005 0 -0.005 -0.01 -0.015 0
(b)
20
60
40
80
100
Age
Figure 3 (a) Net migration rate, Hennepin County, Minnesota, 1990 (combined sexes). y-axis indicates number of net migrants per person-years. (b) Net migration rate, rest of Minnesota, 1990 (combined sexes). y-axis indicates number of net migrants per person-years. Please note the change in y-axis scale as compared to (a).
1244
G.J. Beilman et al.
Figure 4 (a) Hennepin County population, 1970—2024. Median projected Hennepin County population in 2024: approximately 1.3 million, indicating population growth of 17%. Upper and lower 95% confidence intervals: 1.36 and 1.25 million, respectively. Note the non-zero origin of the y-axis. (b) Rest of Minnesota population, 1970—2024. Median projected Minnesota population (excluding Hennepin County) in 2024: approximately 5 million, indicating population growth of 31%. Upper and lower 95% confidence intervals: 5.08 and 4.79 million, respectively. Please note the change in y-axis scale as compared to (a) as well as its non-zero origin.
are projected to grow steadily throughout the forecast period, although growth in Hennepin County slows toward the end of the period. This median forecast and the 95% confidence interval estimations are displayed in Figure 4a and b. Hennepin County is expected to grow 17% over the 25-year period in the median forecast, while the rest of Minnesota is expected to grow 31%. Minnesota as a whole is expected to grow 28% over the 25-year period.
Trauma volume Estimating current trauma rates based on age Trauma volume at our centre from the years 1994 to 1999 was available, and trauma rates were developed using Census Bureau data.12 The average trauma rates by age for Hennepin County are shown in Figure 5a. There is an obvious increase in the trauma rate for both Hennepin County females and males after age 60. A noticeable hump occurs at approximately age 20 for both males and females. This
trauma rate increase in Hennepin County males is exaggerated compared to Hennepin County females. For the rest of Minnesota, the pattern is similar to that observed in Hennepin County (Figure 5b). For both sexes, there is an increase around age 20 as well as a steady increase in trauma rates at older ages. Forecasting future trauma rate Trauma rates of admission to our centre calculated from the 1994 to 1999 data showed an overall upward trend for the 6 years of observation. Based on this data, our model would predict further unchecked increases in trauma rates, producing enormous figures. Because of this, we placed upper boundaries for a maximum rate and lower boundaries for a minimum rate to keep the values from changing too greatly. An upper boundary was set at 125% of the maximum observed rate for a particular age group, and the lower boundary was set at 80% of the minimum observed.
Population-based prediction of trauma volumes
1245
Figure 5 (a) Trauma rates by age, Hennepin County males and females, 1994—1999. y-axis indicates yearly number of admissions to our trauma centre per 10,000 population. (b) Trauma rates by age, rest of Minnesota males and females, 1994—1999. y-axis indicates yearly number of admissions in a year per 10,000 population. Please note the change in yaxis scale as compared to (a).
Forecasting total trauma volume based on future population When applied to the population forecasts, the trauma rate figures produce the trauma volume forecast in Figure 6. According to this forecast, trauma volume is projected to increase substantially. The median forecast estimates that
trauma volume will increase by 57% with confidence intervals estimating a range from 32 to 87%. Figure 7 compares the age distribution of trauma counts in 2000 to the median distribution in 2024. There is a significant shift away from the young and middleaged adult years towards the elderly as the babyboomers move into these ages.
Figure 6 Trauma volume (number of admissions per year) at our Level 1 trauma centre projected to 2024. y-axis indicates number of trauma admissions to our centre. Median increase in trauma volume of 57% between 1999 and 2024. Upper and lower 95% confidence intervals: 32 and 87%, respectively. Please note the non-zero origin of the y-axis.
1246
G.J. Beilman et al.
Figure 7 Distribution of trauma by decade of life, 2000 and 2024. Each bar indicates proportion of total trauma cases per decade of life. Please note that year 2000 is to the right and year 2024 is to the left.
Discussion Using stochastic methods to predict changes in the demographic make-up of populations, the trauma volume at our Level 1 trauma centre is projected to increase 57% by 2024 (approximately 2% per year). A substantial part of this increase stems from a rise in age-specific trauma admission rates as the baby-boom population ages. Additionally, the projected growth in trauma volume is derived from a projected population expansion outside Minneapolis. This would result in increased referrals to our tertiary trauma centre. Another obvious concern is that much of the future trauma volume is predicted to come from the elderly. With worsened outcomes for trauma in the elderly, there is potential for an increase in the number of traumarelated deaths in the future due to this expanding population.12 There is also potential for increased resource requirements as the elderly generally remain hospitalized for a longer period of time after injury.4 A rise in rural elderly trauma is also concerning in light of traditionally poorer rural trauma outcomes.17 At least one third of geriatric trauma patients do not return to pre-injury function.4,13,18 This places a large economic burden on society and the families of these patients as they frequently require subacute care and skilled nursing facilities. Richmond et al. retrospectively evaluated 10 years of trauma outcomes in Pennsylvania and demonstrated that 25.4% of all trauma admissions age 65 and over were
discharged to a skilled nursing facility.16 Actually, this figure may be falsely low as many more elderly patients with less severe injuries are admitted, and therefore able to be discharged to home. In Minnesota, there are very few long-term rehabilitation facilities at present, all with lengthy waiting lists. The potential for a greater number of patients needing these facilities is projected to increase. Therefore, a significant need for more of these centres is likely in the near future. Also clear is the demand for skilled professionals to staff these facilities. Another important point concerns the care of geriatric trauma patients prior to arrival at a trauma centre. Providers of pre-hospital and rural medical care should make transport decisions after fully assessing the physiologic reserve and co-morbidities of the geriatric patient. Initially, the injured geriatric patient may appear to have only mild to moderate injuries, only to succumb to later complications. This is particularly the case in geriatric patients after a fall, the most common type of geriatric trauma admission.16 It has been shown in other series that seriously injured geriatric patients display an increase in survival when triaged using a formal trauma system.11 Therefore, in both urban and rural situations, older patients should be ‘‘over-triaged’’ to a designated trauma centre, as the danger for complications and late morbidity exists.20 If such a justifiably conservative approach is taken, then this is another potential factor driving future trauma admissions at our centre.
Population-based prediction of trauma volumes
There are potential weaknesses of this study. The first includes our use of data at only one trauma centre in the metropolitan area. The use of one centre’s trauma admissions as a sweeping rate may not be an accurate estimate of county/regional rates. Additionally, the 5-year baseline is somewhat short to make longer-term predictions. Another weakness is the method of prediction. Stochastic methods have shortcomings when modelling from a small time series with a definite trend. In such a situation, the stochastic method will forecast a short-term trend into a long-term trend. Placing boundaries on the model can alleviate this problem, however, and the stochastic method still has the advantage of producing trajectories that are more realistic and variable. The validity of the confidence intervals in stochastic forecasting is dependent upon the assumptions underlying the model. In using these models, one should remain cognizant of these assumptions. Finally, when projecting trauma rates, it is difficult to predict the effects of other issues that influence trauma rates, such as fluctuations in the economy or wars, both of which have the potential to significantly change trauma rates for specific demographic groups. Although this study utilized information exclusive to Minnesota, the increase in trauma volume likely forecasts a national trend. As Minneapolis is representative of an average urban setting, this model may be useful for similar cities to forecast future trauma volumes. In light of these findings, the need for emergency medical personnel, trauma surgeons, nurses, and other health-care professionals is evident. Additionally, there is a definite need to foster interest in trauma care for the future. Governmental resources may be required in the form of educational reimbursements or other financial incentives in order to adequately care for this growing population of patients.
Conclusions In conclusion, our analysis indicates that the use of stochastic projections is promising for local demographics and may be useful for forecasting rates of other types of disease. If projections via this method hold true, our Level 1 trauma centre can anticipate a marked increase of 57% in trauma admissions by 2024, and resource planning can be shaped accordingly. There will be a need for additional facilities, emergency medical personnel, trauma surgeons, nurses, and other health-care personnel to adequately care for these injured patients. Based on these data, similar urban areas may also expect such growth in trauma admission rates.
1247
Acknowledgements Financial support for this research is provided by NICHD Training Grant #HD07275-17 and the Office of Naval Research Grant #N00014-02-10093.
References 1. Baker SP, O’Neill B, Ginsherm MJ. The injury fact book. 2nd ed. New York: Oxford University Press; 1992. 2. Baker SP, Whitfield RA, O’Neill B. Geographic variations in mortality from motor vehicle collisions. NEJM 1987;316: 1384—7. 3. Bushby J, Carayvasos GL. Immune function, autoimmunity and selective immunoprophylaxis in the aged. Med Clin North Am 1985;69:465—74. 4. DeKeyser F, Carolan D, Trask A. Suburban geriatric trauma: the experiences of a Level 1 trauma center. Am J Crit Care 1995;4(5):379—82. 5. Internal Revenue Service website http://www.IRS.gov. 6. Lee RD. Modeling and forecasting the time series of US fertility. Int J Forecasting 1993;9:187—202. 7. Lee RD, Carter LR. Modeling and forecasting US mortality. J Am Stat Assoc 1992;87:659—75. 8. Lee RD, Tuljapurkar S. Stochastic forecasts for social security. In: Wise DA, editor. Frontiers in the economics of aging. Chicago: University of Chicago Press; 1998. p. 393—420. 9. Lee RD, Tuljapurkar S. Stochastic population forecasts for the US. J Am Stat Assoc 1994;89:1175—89. 10. MacKenzie EJ, Fowler CJ. Epidemiology. In: Mattox KL, Feliciano DV, Moore EE, editors. Trauma. 4th ed. New York City: McGraw-Hill; 2000. p. 21—40. 11. Mann NC, Cahn RM, Mullins RJ, et al. Survival among injured geriatric patients during construction of a statewide trauma system. J Trauma 2001;50:1111—6. 12. Minnesota Census Bureau statistics US Census bureau website http://www.census.gov. 13. Oreskovich M, Howard J, Copass M, Corrico CJ. Geriatric trauma: injury patterns and outcomes. J Trauma 1984;24: 565—72. 14. Osler T, Hales K, Baack B, et al. Trauma in the elderly. Am J Surg 1988;156:537—43. 15. Perdue PW, Watts DD, Kaufmann CR, Trask AL. Differences in mortality between elderly and younger adult trauma patients: geriatric status increases risk of delayed death. J Trauma 1998;45(4):805—10. 16. Richmond TS, Kauder D, Strumpf N, Meredith T. Characteristics and outcomes of serious traumatic injury in older adults. J Am Geriatr Soc 2002;50:215—22. 17. Rogers FB, Osler TM, Shackford SR, et al. A populationbased study of geriatric trauma in a rural state. J Trauma 2001;50:604—11. 18. Ross N, Timberlake GA, Rubino LJ, Kerstein MD. High cost of trauma care in the elderly. Southern Med J 1989;82(7): 857—9. 19. Ruggles S, Sobeck M. Historical Census Projects: integrated public use microdata series, Version 2.0. Minneapolis: University of Minnesota; 1997. 20. Sattin RW, Mullins RJ. Geriatric trauma: the continuing epidemic. J Am Geriatr Soc 2002;50:394—5. 21. Waller JA. Emergency health services in areas of low population density. JAMA 1969;207:2255—8.
Population-based prediction trauma volumes at a Level 1 trauma centre Greg J. Beilmana,b,*, Jodie H. Taylorb,c, Lisa Joba, Jesse Moend, Aaron Gullicksone a
Department of Surgery, North Trauma Institute, North Memorial Medical Center, Robbinsdale, MN, USA Department of Surgery, Division of Surgical Critical Care, University of Minnesota, MMC 11, 420 Delaware Street SE, Minneapolis, MN 55455, USA c Department of Surgery, Hennepin County Medical Center, Minneapolis, MN, USA d Arithmancer Demographic Consulting, Minneapolis, MN, USA e Department of Demography, University of California, Berkeley, CA, USA b
Accepted 23 March 2004
KEYWORDS Demographics; Elderly; Stochastic; Injury
Summary Objective: With an ageing US population, the demographics of traumatic injuries are being significantly altered. Census projections predict that the number of Americans over age 65 will double in the next 20 years. We used stochastic methods to forecast trauma admissions in order to predict the effects of such demographic changes at our trauma centre. Methods: Age- and sex-related rates of traumatic admission were determined using population statistics and trauma registry data from 1994 to 1999. These rates were then projected from 2000 to 2025 based on both the Lee—Carter and random walk with drift methods. Stochastic population projections were made and paired with the projected trauma rates, allowing estimation of total trauma volume. Results: Trauma rates were predicted to increase for most age groups. Trauma admissions are predicted to increase 57% by 2024. By 2019, 50% of trauma admissions will be 60 or older. Conclusions: Our trauma volume is expected to increase 57% by 2024, an increase of 2% per year. More of this volume will consist of elderly patients, potentially requiring increased health-care resources. ß 2004 Elsevier Ltd. All rights reserved.
Introduction Trauma is traditionally a disease of the young. In patients who are less than 45 years of age, trauma *Corresponding author. Tel.: þ1-612-625-7911; fax: þ1-612-626-0439. E-mail address: [email protected] (G.J. Beilman).
remains the leading cause of death in the United States. Although trauma is only the seventh leading cause of death in patients 65 and older, the death rate from trauma actually increases with age. In these older patients, the rate of injury-related death is 103 per 100,000 population, and for patients aged 75 and older, the injury death rate climbs to 161 per 100,000 population, three times higher than the rate
0020–1383/$ — see front matter ß 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.injury.2004.03.018
1240 for all ages combined.10 Furthermore, time spent in the hospital for similar injuries is two times greater in patients older than 65 than those under 45, mainly because of the increased incidence of complications.10,14,15 This drives an enormous increase in resource utilization for these older patients. The ageing of the US population due to the ‘‘baby-boom’’ generation may soon place a great strain on the capacity of some trauma centres to provide adequate and reliable trauma care. It has been shown that elderly patients over age 70 are almost five times more likely to die from traumatic injury than younger patients even when corrected for severity of injury.3 These deaths typically occur later after injury.3 This is believed to be related to a decline in immunologic and physiologic function and an increased prevalence of pre-existing disease.3,17 Furthermore, in Minnesota, approximately 20% of the elderly population lives in rural areas where there is no access to a Level 1 trauma centre.12 Rural patients who are severely injured require transfer to one of the three Level 1 trauma centres in the neighbouring cities of Minneapolis and St. Paul. It has been shown in other studies that rural patients are more likely to die from traumatic injury than their urban counterparts.1,2,17,21 As the population of Minnesota ages, the number of elderly patients in rural areas is likely to increase, even when accounting for migration. If there is a potential for rural geriatric patients to do poorly when compared to urban geriatric patients, then there may soon be a need for increased availability of trauma care to these patients. There are statistical techniques available to predict future population changes. A stochastic technique has been applied to projections in larger populations such as the United States.7—9 We used this technique along with trauma admission data from the past 5 years to forecast future changes in trauma volume at our Level 1 trauma centre. This project was devised in order to assist in planning for trauma care in the next decades.
Materials and methods Prediction of population trajectories The cohort component method, a common method in demography, was used to project the overall population and its age distribution. This method uses estimates of future fertility, mortality, and migration rates to project birth-cohorts forward in time and to replace them as new birth-cohorts enter the population. In order to incorporate the uncertainty of these future vital rates, a stochastic
G.J. Beilman et al.
method was used. Using this modelling technique, different rates of such variables are sampled from a probability distribution, and a large number of possible population trajectories are created. Stochastic forecasts predict rates which fluctuate realistically, and therefore do not artificially force populations toward a stable configuration.9 This is particularly important when the age distribution of the population matters, as it does in this case. Further, stochastic methods allow values such as the median and 95% confidence interval to be calculated. This model was applied to forecast both Hennepin County (i.e. Minneapolis) and the rest of Minnesota separately in order to predict both local and regional trauma volume, as this hospital serves as a site for both metropolitan and state-wide acute care.
Estimation of population Based on the 2000 US Census, the current estimated population of Hennepin County (i.e. urban Minneapolis) is 1.1 million, and the state of Minnesota (including suburban Minneapolis and neighbouring St. Paul) approaches five million total citizens.12 To facilitate estimation of future population size, the components of population were broken into fertility, mortality, and migration. Age-specific fertility and mortality rates were estimated based on birth counts from 1980 to 1998 and death counts from 1970 to 1999, both available from the Minnesota Department of Health.12 United States Census Bureau estimates of mid-year population size by age from 1970 to 1999 were used as the denominator for these rates.12 However, the Census Bureau underestimated growth in Hennepin County during the 1990s requiring correction of these population counts from 1990 to 1999 based on the observed counts of births and deaths by age each year and net migration implied by the 2000 Census population counts. Data on migration variability was gathered from the Internal Revenue Service.5 An age-specific pattern of net migration for both Hennepin County and the rest of Minnesota was developed from the Public Use Microdata Sample of the 1990 US Census.19 A model described by Lee and Carter was used to forecast each component within reasonable bounds.6,7 In order to predict possible trajectories of population changes, it was assumed that the age-specific patterns of fertility and migration would remain the same. Using the stochastic method, 1000 trajectories of fertility, mortality, and migration were generated, and these trajectories were then used to create 1000 population projections. Each of
Population-based prediction of trauma volumes
the forecasts was for 25 years starting from the 1 January 2000 population.
1241
Results Components of the forecasts
Estimation of trauma volume Retrospective data was collected on all trauma patients from our North Memorial Medical Centre trauma registry over a 5-year period from 1 January 1994 to 1 January 1999. Trauma rates of admission to our trauma centre were developed by using US Census Bureau county estimates as our exposure base, corrected for the 1990s.12 The data was further separated into trauma rates by sex, age, and region. Then a random walk with drift method was used to predict future trauma volume. In order to facilitate this forecast, we made two assumptions: (1) trauma rates will not climb higher than 125% of the highest rate or drop lower than 80% of the lowest rate of rates noted between the years 1994 and 1999; (2) percent of trauma seen at each hospital site will not change. The population forecasts created were then applied to the trauma volume forecast to estimate changes in trauma volume.
Fertility rate The historical and projected total fertility rate for both Hennepin County (Minneapolis) and the rest of Minnesota data are graphed in Figure 1a and b. Hennepin County’s fertility rate shows a definite upward trend in the 1980s that is not apparent for the rest of Minnesota. Since forecasts using this trend would have resulted in an exceedingly high fertility rate in the long-term, it was decided to project ‘‘trendless’’ fluctuation in this case. The median projections for both Hennepin County and the rest of Minnesota are based on 1000 iterations and predict that the total fertility rate will average out at approximately two births per woman over her reproductive life-cycle in both regions, with an upper and lower 95% confidence interval of 1.6 and 2.3 in 2024. These confidence intervals are somewhat smaller than those for the United States as a whole, due to the shorter time series used in this research.
Figure 1 (a) Historic and projected total fertility rate (TFR), Hennepin County, Minnesota. Median projected TFR for 2024 ¼ 2:0 births per woman. Upper and lower 95% confidence intervals ¼ 1:6 and 2.3, respectively. (b) Historic and projected total fertility rate (TFR), rest of Minnesota (excluding Hennepin County). Median projected TFR for 2024 ¼ 1:9 births per woman. Upper and lower 95% confidence intervals ¼ 1:6 and 2.3, respectively.
1242
G.J. Beilman et al.
Figure 2 (a) Female life expectancy, Hennepin County, Minnesota. Median age in 1999 ¼ 82:6 years. Median projected age in 2024 ¼ 87:3 years. (b) Male life expectancy, Hennepin County, Minnesota. Median age in 1999 ¼ 77:6 years. Median projected age in 2024 ¼ 83:2 years. (c) Female life expectancy, rest of Minnesota. Median age in 1999 ¼ 81:5 years. Median projected age in 2024 ¼ 85:0 years. (d) Male life expectancy, rest of Minnesota. Median age in 1999 ¼ 76:4 years. Median projected age in 2024 ¼ 80:9 years.
Life expectancy The Lee—Carter method was used to model the historical mortality time series and a random walk with drift method was used to project mortality
rates into the future.7 The life expectancy forecasts for the different regions were forced to co-vary perfectly in each year such that the differences observed in 1999 between regions and sexes were
Population-based prediction of trauma volumes
1243
maintained. This constraint was added due to the high covariation of the historical time series. The historical and projected life expectancy for Hennepin County females and males is shown in Figure 2a and b, respectively. A median projection based on 1000 iterations is shown which forecasts the anticipated life expectancy for females and males to 2024. There is a definite trend upward in life expectancy for both Hennepin County females and males which continues in the forecast. Life expectancy data for females and males in the rest of Minnesota is shown in Figure 2c and d, respectively. As with the Hennepin County data, the median projection for the rest of Minnesota predicts that both the female and male populations will live longer. Migration Age-specific rates of net migration were estimated from the 1990 Census Public Use Microdata Sample for Minnesota.19 At approximately age 20—24 years, there is a net positive migration for both females and males into Hennepin County (Figure 3a). (This may represent an influx of post-college graduates
entering the urban workforce.) After this large flux, the net migration rate declines and then plateaus in the slight negative range. In the late teen years, the rest of Minnesota sees an increase in outward migration as children leave home (Figure 3b). A large upswing is then evident in the late twenties to thirty age group for both females and males. The net migration for the rest of Minnesota then remains slightly positive until another increase occurs in the elderly population (1970s—1980s). The Internal Revenue Service provides information based on tax returns on total in and out migration at the county level for every county in the United States.5 Combining these data with the age schedules of net migration rates and population estimates produced estimates of the ‘magnitude’ of net migration in a given year. These ‘magnitudes’ were used to estimate future variability in net migration by simple random drawing.
Final population projection Taking all components into consideration, the populations of Hennepin County and the rest of Minnesota
Net Migration Rate, Hennepin County, 1990 (Combined Sexes) 0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 -0.05 0
20
40
(a)
60
80
100
Age
Net Migration Rate, Rest of Minnesota, 1990 (Combined Sexes) 0.015 0.01 0.005 0 -0.005 -0.01 -0.015 0
(b)
20
60
40
80
100
Age
Figure 3 (a) Net migration rate, Hennepin County, Minnesota, 1990 (combined sexes). y-axis indicates number of net migrants per person-years. (b) Net migration rate, rest of Minnesota, 1990 (combined sexes). y-axis indicates number of net migrants per person-years. Please note the change in y-axis scale as compared to (a).
1244
G.J. Beilman et al.
Figure 4 (a) Hennepin County population, 1970—2024. Median projected Hennepin County population in 2024: approximately 1.3 million, indicating population growth of 17%. Upper and lower 95% confidence intervals: 1.36 and 1.25 million, respectively. Note the non-zero origin of the y-axis. (b) Rest of Minnesota population, 1970—2024. Median projected Minnesota population (excluding Hennepin County) in 2024: approximately 5 million, indicating population growth of 31%. Upper and lower 95% confidence intervals: 5.08 and 4.79 million, respectively. Please note the change in y-axis scale as compared to (a) as well as its non-zero origin.
are projected to grow steadily throughout the forecast period, although growth in Hennepin County slows toward the end of the period. This median forecast and the 95% confidence interval estimations are displayed in Figure 4a and b. Hennepin County is expected to grow 17% over the 25-year period in the median forecast, while the rest of Minnesota is expected to grow 31%. Minnesota as a whole is expected to grow 28% over the 25-year period.
Trauma volume Estimating current trauma rates based on age Trauma volume at our centre from the years 1994 to 1999 was available, and trauma rates were developed using Census Bureau data.12 The average trauma rates by age for Hennepin County are shown in Figure 5a. There is an obvious increase in the trauma rate for both Hennepin County females and males after age 60. A noticeable hump occurs at approximately age 20 for both males and females. This
trauma rate increase in Hennepin County males is exaggerated compared to Hennepin County females. For the rest of Minnesota, the pattern is similar to that observed in Hennepin County (Figure 5b). For both sexes, there is an increase around age 20 as well as a steady increase in trauma rates at older ages. Forecasting future trauma rate Trauma rates of admission to our centre calculated from the 1994 to 1999 data showed an overall upward trend for the 6 years of observation. Based on this data, our model would predict further unchecked increases in trauma rates, producing enormous figures. Because of this, we placed upper boundaries for a maximum rate and lower boundaries for a minimum rate to keep the values from changing too greatly. An upper boundary was set at 125% of the maximum observed rate for a particular age group, and the lower boundary was set at 80% of the minimum observed.
Population-based prediction of trauma volumes
1245
Figure 5 (a) Trauma rates by age, Hennepin County males and females, 1994—1999. y-axis indicates yearly number of admissions to our trauma centre per 10,000 population. (b) Trauma rates by age, rest of Minnesota males and females, 1994—1999. y-axis indicates yearly number of admissions in a year per 10,000 population. Please note the change in yaxis scale as compared to (a).
Forecasting total trauma volume based on future population When applied to the population forecasts, the trauma rate figures produce the trauma volume forecast in Figure 6. According to this forecast, trauma volume is projected to increase substantially. The median forecast estimates that
trauma volume will increase by 57% with confidence intervals estimating a range from 32 to 87%. Figure 7 compares the age distribution of trauma counts in 2000 to the median distribution in 2024. There is a significant shift away from the young and middleaged adult years towards the elderly as the babyboomers move into these ages.
Figure 6 Trauma volume (number of admissions per year) at our Level 1 trauma centre projected to 2024. y-axis indicates number of trauma admissions to our centre. Median increase in trauma volume of 57% between 1999 and 2024. Upper and lower 95% confidence intervals: 32 and 87%, respectively. Please note the non-zero origin of the y-axis.
1246
G.J. Beilman et al.
Figure 7 Distribution of trauma by decade of life, 2000 and 2024. Each bar indicates proportion of total trauma cases per decade of life. Please note that year 2000 is to the right and year 2024 is to the left.
Discussion Using stochastic methods to predict changes in the demographic make-up of populations, the trauma volume at our Level 1 trauma centre is projected to increase 57% by 2024 (approximately 2% per year). A substantial part of this increase stems from a rise in age-specific trauma admission rates as the baby-boom population ages. Additionally, the projected growth in trauma volume is derived from a projected population expansion outside Minneapolis. This would result in increased referrals to our tertiary trauma centre. Another obvious concern is that much of the future trauma volume is predicted to come from the elderly. With worsened outcomes for trauma in the elderly, there is potential for an increase in the number of traumarelated deaths in the future due to this expanding population.12 There is also potential for increased resource requirements as the elderly generally remain hospitalized for a longer period of time after injury.4 A rise in rural elderly trauma is also concerning in light of traditionally poorer rural trauma outcomes.17 At least one third of geriatric trauma patients do not return to pre-injury function.4,13,18 This places a large economic burden on society and the families of these patients as they frequently require subacute care and skilled nursing facilities. Richmond et al. retrospectively evaluated 10 years of trauma outcomes in Pennsylvania and demonstrated that 25.4% of all trauma admissions age 65 and over were
discharged to a skilled nursing facility.16 Actually, this figure may be falsely low as many more elderly patients with less severe injuries are admitted, and therefore able to be discharged to home. In Minnesota, there are very few long-term rehabilitation facilities at present, all with lengthy waiting lists. The potential for a greater number of patients needing these facilities is projected to increase. Therefore, a significant need for more of these centres is likely in the near future. Also clear is the demand for skilled professionals to staff these facilities. Another important point concerns the care of geriatric trauma patients prior to arrival at a trauma centre. Providers of pre-hospital and rural medical care should make transport decisions after fully assessing the physiologic reserve and co-morbidities of the geriatric patient. Initially, the injured geriatric patient may appear to have only mild to moderate injuries, only to succumb to later complications. This is particularly the case in geriatric patients after a fall, the most common type of geriatric trauma admission.16 It has been shown in other series that seriously injured geriatric patients display an increase in survival when triaged using a formal trauma system.11 Therefore, in both urban and rural situations, older patients should be ‘‘over-triaged’’ to a designated trauma centre, as the danger for complications and late morbidity exists.20 If such a justifiably conservative approach is taken, then this is another potential factor driving future trauma admissions at our centre.
Population-based prediction of trauma volumes
There are potential weaknesses of this study. The first includes our use of data at only one trauma centre in the metropolitan area. The use of one centre’s trauma admissions as a sweeping rate may not be an accurate estimate of county/regional rates. Additionally, the 5-year baseline is somewhat short to make longer-term predictions. Another weakness is the method of prediction. Stochastic methods have shortcomings when modelling from a small time series with a definite trend. In such a situation, the stochastic method will forecast a short-term trend into a long-term trend. Placing boundaries on the model can alleviate this problem, however, and the stochastic method still has the advantage of producing trajectories that are more realistic and variable. The validity of the confidence intervals in stochastic forecasting is dependent upon the assumptions underlying the model. In using these models, one should remain cognizant of these assumptions. Finally, when projecting trauma rates, it is difficult to predict the effects of other issues that influence trauma rates, such as fluctuations in the economy or wars, both of which have the potential to significantly change trauma rates for specific demographic groups. Although this study utilized information exclusive to Minnesota, the increase in trauma volume likely forecasts a national trend. As Minneapolis is representative of an average urban setting, this model may be useful for similar cities to forecast future trauma volumes. In light of these findings, the need for emergency medical personnel, trauma surgeons, nurses, and other health-care professionals is evident. Additionally, there is a definite need to foster interest in trauma care for the future. Governmental resources may be required in the form of educational reimbursements or other financial incentives in order to adequately care for this growing population of patients.
Conclusions In conclusion, our analysis indicates that the use of stochastic projections is promising for local demographics and may be useful for forecasting rates of other types of disease. If projections via this method hold true, our Level 1 trauma centre can anticipate a marked increase of 57% in trauma admissions by 2024, and resource planning can be shaped accordingly. There will be a need for additional facilities, emergency medical personnel, trauma surgeons, nurses, and other health-care personnel to adequately care for these injured patients. Based on these data, similar urban areas may also expect such growth in trauma admission rates.
1247
Acknowledgements Financial support for this research is provided by NICHD Training Grant #HD07275-17 and the Office of Naval Research Grant #N00014-02-10093.
References 1. Baker SP, O’Neill B, Ginsherm MJ. The injury fact book. 2nd ed. New York: Oxford University Press; 1992. 2. Baker SP, Whitfield RA, O’Neill B. Geographic variations in mortality from motor vehicle collisions. NEJM 1987;316: 1384—7. 3. Bushby J, Carayvasos GL. Immune function, autoimmunity and selective immunoprophylaxis in the aged. Med Clin North Am 1985;69:465—74. 4. DeKeyser F, Carolan D, Trask A. Suburban geriatric trauma: the experiences of a Level 1 trauma center. Am J Crit Care 1995;4(5):379—82. 5. Internal Revenue Service website http://www.IRS.gov. 6. Lee RD. Modeling and forecasting the time series of US fertility. Int J Forecasting 1993;9:187—202. 7. Lee RD, Carter LR. Modeling and forecasting US mortality. J Am Stat Assoc 1992;87:659—75. 8. Lee RD, Tuljapurkar S. Stochastic forecasts for social security. In: Wise DA, editor. Frontiers in the economics of aging. Chicago: University of Chicago Press; 1998. p. 393—420. 9. Lee RD, Tuljapurkar S. Stochastic population forecasts for the US. J Am Stat Assoc 1994;89:1175—89. 10. MacKenzie EJ, Fowler CJ. Epidemiology. In: Mattox KL, Feliciano DV, Moore EE, editors. Trauma. 4th ed. New York City: McGraw-Hill; 2000. p. 21—40. 11. Mann NC, Cahn RM, Mullins RJ, et al. Survival among injured geriatric patients during construction of a statewide trauma system. J Trauma 2001;50:1111—6. 12. Minnesota Census Bureau statistics US Census bureau website http://www.census.gov. 13. Oreskovich M, Howard J, Copass M, Corrico CJ. Geriatric trauma: injury patterns and outcomes. J Trauma 1984;24: 565—72. 14. Osler T, Hales K, Baack B, et al. Trauma in the elderly. Am J Surg 1988;156:537—43. 15. Perdue PW, Watts DD, Kaufmann CR, Trask AL. Differences in mortality between elderly and younger adult trauma patients: geriatric status increases risk of delayed death. J Trauma 1998;45(4):805—10. 16. Richmond TS, Kauder D, Strumpf N, Meredith T. Characteristics and outcomes of serious traumatic injury in older adults. J Am Geriatr Soc 2002;50:215—22. 17. Rogers FB, Osler TM, Shackford SR, et al. A populationbased study of geriatric trauma in a rural state. J Trauma 2001;50:604—11. 18. Ross N, Timberlake GA, Rubino LJ, Kerstein MD. High cost of trauma care in the elderly. Southern Med J 1989;82(7): 857—9. 19. Ruggles S, Sobeck M. Historical Census Projects: integrated public use microdata series, Version 2.0. Minneapolis: University of Minnesota; 1997. 20. Sattin RW, Mullins RJ. Geriatric trauma: the continuing epidemic. J Am Geriatr Soc 2002;50:394—5. 21. Waller JA. Emergency health services in areas of low population density. JAMA 1969;207:2255—8.