- Email: [email protected]
Sentinel health events surveillance in diabetes Deaths among persons under age 45 with diabetes
.J Clin Epidemiol
Vol. 41, No.
IO, pp. 999-1006,
0895-4356/88$3.00+ 0.00 Pergamon Press plc
1988
Printed in Great Britain
SENTINEL
DEATHS
HEALTH EVENTS SURVEILLANCE IN DIABETES AMONG
PERSONS UNDER DIABETES
AGE 45 WITH
DIANE BILD I,* LINDA S. GEE,’ STEVENM. TEUTXH,’ BARBARAGABELLA,~ BARBARAA. HUDSPETH,~RALPH M. SCHUBERT,~CHARLESW. GOLLMAR,’ DOUGLASJ. KAPLAN,~FREDERICKA. CONNELL’and JULIE C. WILL’ ‘Division of Diabetes Control and 2Epidemiology Program Office, Centers for Disease Control, Atlanta, Georgia, 3Colorado Department of Health, 4111inoisDepartment of Health, 5Kentucky Department of Health Services, 6Louisiana Department of Health and Human Resources and ‘Washington Department of Social and Health Services, U.S.A. (Received in revised form 28 March 1988)
Abstract-The pilot study for a sentinel health events surveillance system for deaths among persons under age 45 with diabetes was conducted in six States in 1984 and 1985. Two hundred and thirty-three events were identified. Information from death certificates, physicians, and families revealed that 22% died from acute complications of diabetes and 53% from chronic complications. Blood pressure measurement and urinalysis testing had been performed in the last year for almost all of the decedents, but other preventive practices were reported less frequently. Hypertension was present in 57% and of those, was not controlled in 73%. Forty-four percent were cigarette smokers at the time of death. Agreement between physicians and families was generally higher for clinical conditions than for care practices. This surveillance system appears to yield information about the health care of persons with diabetes not readily available from other sources, although modifications may be necessary before implementation. Diabetes cations
Chronic disease
Surveillance
Mortality
INTRODUCTION
Surveillance is a mechanism to systematically record and track disease in defined populations over time. Its purposes are to measure disease frequency, risk factors, and trends in diseases and risk factors, to generate hypotheses regarding causation and prevention strategies, and to form the basis for population-based control programs. The classic model disease surveillance is employed for infectious diseases, including the reporting and enumeration of cases with
*Reprint requests should be addressed to: Diane Bild, Division of Diabetes Control, Centers for Disease Control, 1600 Clifton Rd NE, Atlanta, GA 30333, U.S.A.
Health care
Diabetes compli-
analysis and dissemination of results on a routine basis. Chronic disease surveillance is more difficult to conduct because the incidence of chronic diseases is difficult to measure and is only one of many endpoints of interest. Disease onset is insidious, there are multiple disease endpoints for a single disease, and diagnostic criteria for disease endpoints are non-uniform and relatively elaborate. Death, perhaps the most commonly recorded chronic disease endpoint, is often a poor reflection of morbidity. At other end of the disease spectrum, the measurement of risk factors for chronic diseases is desirable, because risk factors relate to disease and can be targeted directly for intervention. However, chronic diseases have multiple risk factors and long latency periods between the
999
DIANE BILD et al.
1000
risk factor and disease onset. Despite the heavy toll in health care costs and human suffering from chronic diseases in the United States, there are few routine reporting mechanisms for them. Diabetes was the seventh leading cause of mortality in the United States in 1984 [1] and the twelfth leading cause of years of potential life lost [2]. Sources of data on diabetes include death certificates, periodic population-based surveys, hospital discharge summaries, and clinic-based studies or registries. No single, currently available data source has all the desired features of a surveillance system-timeliness, a community base (as opposed to clinic base), specificity for diabetes, data related to morbidity and health care factors, and systematic feedback to practitioners in the community. While a variety of data sources are necessary for chronic disease control, the surveillance system for sentinel health events in diabetes described in this paper is one system with potential to address these requirements. A sentinel health event has been described by Rutstein et al. as the occurrence of unnecessary disease, unnecessary disability, or untimely death [3]. These events, designated because of their potential to be prevented, may serve as a reflection of the quality of health care in the community and as a warning signal that the quality of care may need to be improved. Under the assumption that mortality is most preventable in the younger population, we have chosen to focus on persons who died under age 45 with diabetes. A surveillance system was designed to (1) identify persons who died under age 45 with diabetes and (2) assess morbidity and health care practices in this group. This paper describes the pilot study for this surveillance system which was conducted in six States. METHODS
Six State health department-based Diabetes Control Programs (DCP) participated in the pilot study (Colorado, Illinois, Kentucky, Louisiana, Michigan, and Washington). Death certificates were collected for periods of 4-8 months from July 1984 through July 1985. It was not possible to assure that all months of the year were represented equally among States. An event was defined as any death of a person under age 45, with diabetes mentioned anywhere on the death certificate. Using standardized data collection instruments, DCP investigators gathered basic demographic data
from death certificates and, when feasible, conducted follow-back surveys of both physicians associated with the decedent and family members to obtain clinical and health care-related information. Death certificate information was collected as recorded on the certificate. Followback information was obtained by mail, by telephone, or through face-to-face interview. The form contained 16 questions. Questions were phrased as appropriate to the understanding of either medical personnel or laypersons. For example, “myocardial infarction” was used for medical personnel and “heart attack” was used for laypersons. The presence or history of clinical conditions and use of care practices in the last year of life were ascertained through a “yes”, “no”, or “unknown” response. Hypertension was defined as blood pressure greater than 140/90 mmHg. Legal blindness was defined as visual acuity of 20/200 or poorer. Nephropathy was defined as persistent proteinuria in the absence of infection. Other questions asked were type of medication for diabetes (insulin, oral agent, both, neither, or unknown), whether blood pressure in the last year was 2 140/90 mmHg, height, weight, smoking habits (never smoked, ex-smoker, smoked< 1 pack per day, smoked > 1 pack per day, or unknown), the level of compliance with pharmacologic and non-pharmacologic therapies (excellent, average, poor, or unknown), and the average number of times per year the decedent had seen a physician. Using all information available, the cause of death in each event was classified by two of the authors (D.B. and L.G.) into one of three categories based on the one condition most contributory to death. The categories included those attributed to (1) acute complications of diabetes (diabetic ketoacidosis or coma, infection, and hypoglycemia), (2) chronic complications (heart disease, end-stage renal disease, and cerebrovascular disease), and (3) other causes (diabetes only, not diabetes-related, and unknown). The mortality rate for each State was calculated by using the number of deaths (from death certificates) in the numerator and national diabetes prevalence data by region (National Health Interview Survey, NCHS, 1980) and the time period (fraction of a year) in the denominator. The expected rate among all States was calculated based on 1982 national death certificate data [4]. Chi-square analysis was per-
1001
Diabetes Sentinel Health Events Surveillance Table 1, Cause of death among 233 persons under 45 with diabetes, by age group Age in years Cause of death Acute complication Chronic complication Diabetes only, unrelated, or unknown Total @‘I
t&24 WI
25-34 (%I
3544 W)
Total W)
50.0
22.1
18.4
21.9
11.1
48.5
60.5
53.2
38.9
29.4
21.1
24.9
100.0 (18)
100.0 (68)
100.0 (147)
100.0 (233)
formed for comparison of categorical variables including demographic characteristics, clinical conditions, and health care practices. The Wilcoxon rank sum test was used to compare differences in distributions in variables which were not normally distributed. The degree of agreement between physician and family sources on survey data items was calculated using the kappa statistic, an index of agreement corrected for chance-expected agreement [5]. RESULTS
Two hundred and thirty-three events were identified in six States. Assuming the periods of data collection used by the States were representative of the entire year, the average observed annual mortality rate was 32.7 per 10,000 persons under age 45 with diabetes. The expected rate was 34.8 per 10,000 persons under age 45 with diabetes. The death certificate was the only source of information for 37% of the events. Follow-back information from physicians was available for 49% of the events and from families for 35% of the events. Both forms of follow-back information were available for 22% of the events. Decedents included 146 males and 87 females. The male to female mortality ratio was 1.78, compared to 1.93 in the general population under age 45 (based on 1984 mortality statistics for the United States). Twenty-two percent of the decedents died from acute complications of diabetes, 53% died from chronic complications, and 25% died from “diabetes only” or unrelated or unknown causes (Table 1). The distributions of cause of death were significantly different among the three different age groups (x2 = 17.9, p < 0.01). In the
youngest group (0-24yr) 50% of the deaths were due to acute complications, while in the oldest group (35-44yr) 18% died from acute complications. In the youngest age group 11% died from chronic complications, while in the oldest age group 60% died from chronic complications. There was no significant difference in the distributions of cause of death between males and females (Table 2). The median duration of diabetes for those who died from chronic complications was significantly longer than for those who died from acute complications (median duration 23 as compared to 14 yr, Wilcoxon rank sum test, p < 0.01). Several conditions were recorded with notable frequency on death certificates or among “other” conditions in the follow-back surveys. These included alcohol abuse (13 events), suicide (5 events), and cardiomyopathy or congestive heart failure in the absence of ischemic heart disease (7 events). Among the 13 events in which alcohol abuse was reported, 8 of the decedents were male; 5 of the 13 died from acute complications, 3 from chronic complications, and 5 from other causes or diabetes only. In the 5 events in which suicide was the cause of death, all of the decedents were male. The degree of agreement between sources of follow-back information (kappa statistic) was greatest for the presence of the clinical conditions dialysis or transplantation (1 .OO), nephropathy (0.74), retinopathy (0.73) and lower extremity amputation (0.73) [Table 3(a)]. It was also high for whether the decedent had had a blood pressure check in the last year (0.79). Agreement was poorer for survey items on laboratory testing, funduscopic examination, and self care practices. There was poorest agreement on the decedent’s compliance with nonpharmacologic and pharmacologic therapy (0.13 and 0.14, respectively) [Table 3(b)]. The kappa statistic may be considered to reflect agreement beyond that expected by chance as
Table 2. Cause of death among 233 persons under age 45 with diabetes, by sex Cause of death
Male W)
Female WI
Acute Chronic Unknown/diabetes Total WI
20.6 54.1 25.3 100.0 (146)
24.1 51.7 24. I 100.0 (87)
only
Table 3(a). Degree of agreement between physicians and families on clinical condition and care practices in the last year of life among persons with diabetes who died under age 45
Condition or practice Retinopathy Legal blindness Nephropathy Dialysis or transplant Lower extremity amputation Stroke Myocardial infarction History of hypertension Urinalysis in last year Glycohemoglobin test in last year Funduscopic examination in last year Blood pressure check Home glucose monitoring
Number of paired responses
Percent “yes” from family
Percent “yes” from physician
Agreement (kappa statistic)
43 41 41 43 41 38 43 44 36
67.4 36.6 61.0 37.2 17.1 10.5 30.2 52.3 94.4
69.8 39.0 68.3 37.2 14.6 5.3 18.6 65.9 94.4
0.73 0.64 0.74 1.00 0.73 0.64 0.60 0.63 0.46
18
77.8
38.9
0.31
2 39
58.3 95.0 69.2
80.6 92.5 53.8
0.26 0.79 0.26
There was a total of 50 events in which both sources of information were available. The number of paired responses excludes unknown and missing data. Table 3(b). Degree of agreement between physicians and families on clinical condition and care practices in the last year of life among persons with diabetes who died under age 45 Number of paired responses
Condition or practice Medication for diabetes
Family and physician responses Family
46
insulin oral both neither
Agreement (kappa statistic) 0.70
82.6 4.3 6.5 6.5
Physician
insulin oral both neither Cigarette smoking
84.8 6.5 4.3 4.3 0.52
40 Family
never ex-smoker current
57.5 22.5 20.0
Physician
never ex-smoker current Compliance-non-drug
therapy
60.0
10.0 30.0 0.13
36 Family
excellent average poor
33.3 44.4 22.2
Physician
excellent average poor Compliance-drug
therapy
16.7 38.9 44.4 0.14
37 Family
excellent average poor
70.3 24.3 5.4
Physician
excellent average poor
32.4 40.5 27.0
There was a total of 50 events in which both sources of information were available. The number of paired responses excludes unknown and missing data. 1002
1003
Diabetes Sentinel Health Events Surveillance Table 4. Prevalence of clinical conditions among persons with diabetes who died under age 45, by cause of death Cause of death Acute
Unknown/ diabetes only
Chronic
Total
Condition
N
% Positive responses
N
% Positive responses
N
% Positive responses
N
% Positive responses
Retinopathy Blindness Nephropathy Dialysis or transplant Amputation Stroke Myocardial infarction Hypertension
26 21 28
50.0 22.2 35.7
69 64 69
81.2 42.2 72.5
31 35 34
46.0 8.6 35.3
132 126 131
65.2 28.6 55.0
29 30 30
17.2 16.1 6.1
68 69 65
50.0 21.7 12.3
36 36 35
5.6 5.6 2.9
133 135 130
30.8 16.3 8.5
28 26
7.1 23.1
70 73
47.1 74.0
35 33
11.4 45.4
133 132
29.3 56.8
Prevalence of the various factors was calculated using physician responses first, and family responses if no physician response was available. N is the total number of responses excluding unknowns.
follows: values greater than 0.75 represent excellent agreement beyond chance, values below 0.40 represent poor agreement beyond chance, and values between 0.40 and 0.75 represent fair to good agreement beyond chance [S]. Families assessed decedents as having less nephropathy, more heart attacks (compared to myocardial infarctions, the term used for physicians), less high blood pressure (compared to hypertension, the term used for physicians), and using home glucose monitoring more than did physicians. They also tended to rate patients better than did physicians on compliance with both drug and non-drug therapies. Among family respondents the response concerning glycohemoglobin testing (described also as “blood test which determines blood sugar level over a long period of time”) included mostly unknowns. The prevalence of diabetic complications was generally high, as shown in Table 4. Among the most frequent complications were retinopathy (65%), nephropathy (55%), end-stage renal disease (31%) and hypertension (57%). The prevalence was higher among those who died from chronic complications than among those who died from acute complications. Hypertension was not controlled (reported as greater than 140/90 mmHg in the last year) in 73% of those who were reported to have had hypertension (Table 5). Almost all of the decedents had had blood pressure measurement and a urinalysis in the last year of life (Table 6). Among the 87% of the decedents who used insulin, 44% had had a glycohemoglobin measured in the last year and 50% had used self-monitoring of blood glucose.
Almost 44% were smokers at the time of death. While 79% had had an ophthalmic exam in the last year, it had been performed by an ophthalmologist in only 48% of these cases. DISCUSSION
It has been estimated that, with currently available technology for diabetes care, between 50 and 85% of complications of diabetes are treatable and preventable [6]. For example, it has recently been shown that the rate of decline in renal function may be slowed approximately 60% by aggressive antihypertensive management in selected patients with diabetes [7,8]. Previous studies on premature mortality among persons with diabetes have found a significant proportion of the deaths were due to preventable factors. In Washington State, the mor-
Table 5. Hypertension history and treatment, blood pressure measurement, and hypertension control among persons with diabetes who died under age 45
Hypertension history Hypertension history and under treatment Hypertension history and blood pressure checked in last year of life Hypertension history and blood pressure > 140/90 mmHg
N
% Positive responses
132
56.8
54
90.7
56
100.0
41
13.2
Prevalence of each condition was calculated using physician responses first, and family responses if no physician response was available. N is the total number of responses excluding unknowns.
1004
DIANE BILD et al.
Table 6. Care practices by sex in the last year of life among persons with diabetes who died under age 45 Female
Male
Blood pressure check Urinalysis Current smoker Glycohemoglobin test among insulin users Self-monitoring of blood glucose among insulin users Funduscooic examination Funduscopic examination performed by an ophthalmologist Median number of physician visits per year
Total
N
% Positive responses
N
% Positive responses
N
% Positive responses
61 61 75
91.8 86.9 44.0
38 37 42
100.0 100.0 42.9
99 98 117
95.0 91.8 43.6
53
39.6
26
53.8
79
44.3
64 63
48.4 76.2
38 36
52.6 83.3
102 99
50.0 78.8
28 N 79
39.2 Visits
18 -N
61.1 Visits
46 N
47.8 Visits
8
45
7
124
8
Prevalence of the various factors was calculated using physician responses first, and family responses if no physician response was available. N is the total number of responses excluding unknowns.
tality rate for persons under age 45 with diabetes has been reported to be eight times that of the general population under age 45 [9]. A detailed review of death certificates among persons under age 45 with diabetes found almost one-third of the deaths were due to acute complications for which there is definitive therapy. In a study in Great Britain, “neglect of diabetes” contributed to 27% of the deaths among 447 persons with diabetes who died under age 50 [lo]. Thus, it appears that improvements in health care may have an impact on the population of persons with diabetes under age 45. An examination of the morbidity and health care of the segment of the group that died may provide clues to how health care could be improved. Rates of complications of diabetes among the decedents in this study were very high. Over half had diabetic nephropathy, a condition associated with a 3 to 4-fold excess mortality rate [ 111. Clinical intervention (e.g. improved glycemic control and hypertension control) are necessary at least several years prior to the time of death to alter the course of this disease. Younger decedents were at greater risk from acute complications than older decedents. This may be due to a combination of death due to ketoacidosis at diagnosis, poor self-management of insulin, and the fact that chronic complications take more time to develop. Rates of such routine care practices as measurement of blood pressure and urinalysis testing were high, although one would expect to find that all persons with diabetes had had such basic exams, especially during the last year of
life. At this time health care needs were great because of clinically manifest complications of diabetes. However, while blood pressure was being measured routinely, hypertension was not controlled in a substantial proportion of these persons. The rates of glycohemoglobin testing and self blood glucose monitoring among those on insulin were relatively low (44 and 50%, respectively). This may be a reflection of suboptimal quality of care, as one might expect that virtually all persons on insulin use these technologies, especially those under 45. The prevalence of smoking of 44% was higher than the prevalance of 28% for adult males and 25% for adult females in the general population reported in 1985 [12]. The combination of alcohol and diabetes appears to be a dangerous combination. Alcohol has well-observed adverse effects on the health of persons with diabetes, including metabolic disturbances and compromise of the discipline and self-care integral to overall diabetes care [13]. A modest seasonal trend in diabetes mortality has been described with higher rates occurring during the winter months [4]. This may have had a minor effect on the calculated State mortality rates in this study, since States only monitored deaths for part of the year. Ideally, a surveillance system which seeks information on morbidity would obtain information from sources of health care and/or recipients of health care-hospitals, outpatient care centers, and community surveys. Unfortunately, data from these sources are not uniformly available, systematized, linked, or
1005
Diabetes Sentinel Health Events Surveillance
timely. As demonstrated in this pilot project, the death certificate provides a uniformly available, standardized means of identifying persons with diabetes, may be used to obtain data on mortality rates and trends directly, and can serve as an entry point to retrospectively obtain morbidity data. Death among persons under age 45 with diabetes is just one possible event which may be used as an indicator of the overall quality of diabetes care. Diabetic ketoacidosis and hypoglycemia are considered highly treatable with available technology and preventable with adequate monitoring and education; therefore deaths due to acute complications have been suggested as such a measure [3,9]. Other specific causes of death might also be monitored, such as death due to end-stage renal disease. In this pilot study the younger segment of the diabetic population was targeted to increase the proportion of preventable deaths represented among the events, because the risk of the complications of diabetes and mortality increase with age. However, it would be expected that remediable problems discovered in the health care for the diabetic population under age 45 also exist in the rest of the diabetic population. While in the present study each death was considered a sentinel health event, this is probably only appropriate for conditions for which the expected mortality rate is zero. For diabetes surveillance, death rates may be more appropriate units of measurement. Rates may be monitored, and if an unusually high rate is detected, data on decreased individuals may then be collected. It would be desirable to have established expected rates of mortality and normative standards for complications and care practices before launching follow-back studies. Surveillance information must be communicated back to local communities and used to further investigate possible problems, e.g. conduct case-control studies or more intensive monitoring of a suspected problem in a defined region. New systems for surveillance must be evaluated for the sensitivity of the system, the quality of data generated, the potential uses of the system, and the feasibility of implementation. The actual sensitivity of the system for identifying events is probably not as important as that it remains relatively constant. The sensitivity of this system in identifying appropriate events will depend on the completeness of recording of diabetes of death certificates and ascertainment
of death certificates, and on the reliability and completeness of follow-back survey information. Diabetes is listed on death certificates for only about half of the deaths among persons with diabetes [14], and this may vary by State and over time. Diabetes is more often listed as the underlying cause of death in younger persons than older persons with diabetes [15]. It may be assumed that the probability of diabetes being listed anywhere is also greater at younger ages. Thus, the use of a relatively young group of persons in the diabetic population increases the sensitivity of the system. In a related project in Michigan conducted during and after this pilot study, 77 of 84 events (91.7%) identified from both death certificates and hospital records were identified from death certificates alone. In seven of the 84 events there was no mention of diabetes on the death certificate (M. Halpern, personal communication). This provides some evidence that death certificates provide a fairly high degree of sensitivity in identifying these events. The major potential limitation of this system is the labor-intensiveness-the complexity and the time involved (between 4 and 6 hr per completed investigation in one State). Identification of events was relatively easy, but the follow-back involved active information tracking and retrieval. In the implementation of this system some survey data will be deleted and several options for reducing the burden of investigation have been proposed. Options for the latter include conducting follow-back investigation only when an unusual rate of events has been identified, investigating only a subsample of events, limiting sources of followback information to either physicians or families, and limiting the definition of an event to more specific diagnoses, such as death due to diabetic ketoacidosis. Further experience in sentinel health event surveillance will allow this system to be a useful instrument of investigation and, ultimately, to improve the health care system. Acknowledgements-We wish to thank Martin Halpern, PhD. John Beaslev. MA. and Steven Lonaabauah. MS. Michigan Department of Public Health, who gathered and analyzed data in Michigan and assisted in the preparation of this manuscript.
REFERENCES 1.
National Center for Health Statistics. Advance report of final mortality statistics, 1984. Monthly Vital Statistics Report. Hyattsville, Md.: Public Health Service; 26
1006
2.
3.
4.
5. 6.
I.
8.
DIANE BILD er al.
September 1986: Vol. 35, No. 6, Suppl. 2. DHHS Pub No. (PHS) 861120. CDC. Premature mortality from diabetes-use of sentinel health event surveillance to assess causes. MMWR 1986; 35: 711-714. Rutstein DD, Berenberg W, Chalmers TC et al. Measuring the quality of medical care: a clinical method. N Enel J Med 1976: 294: 582-588. National Cent& for Health Statistics. Advance report of final mortality statistics, 1982. Monthly Vital Statistics Report. Hyattsville, MD.: Public Health Service; 20 December 1984; Vol. 33, No. 9, Suppl. 2. DHHS Pub. No. (PHS) 85-1120. Fleiss JL. Statistical Methods for Rates and Proportions. New York: John Wiley; 1981: 2nd edn, 221-225. Carter Center. Closing the gap: The problem of diabetes in the United States. Diabetes Care 1985; 8: 391406. Mogensen CE. Long-term antihypertensive treatment inhibiting progression of diabetic nephropathy. Br Med J 1982; 285: 685688. Parving HH, Andersen AR, Smidt UM, Svendsen PA. Early aggressive antihypertensive treatment in kidney function in diabetic nephropathy. Lancet 1983; 1: 1175-l 178.
9.
10.
Il.
12.
13.
14.
15.
Connell FA. Epidemiologic approaches to the identification of problems in diabetes care. Diabetes Care 1985; 8(Suppl. 1): 82-86. Tunbridge WMG. Occasional Survey: Factors contributing to deaths of diabetics under fifty years of age. Lancet 1981: 569-572. Deckert T, Paulsen JE, Larsen M. Prognosis of diabetics with diabetes onset before the age of thirtyone. 1. survival, causes of death, and complications. Diabetologia 1978; 14: 363-370. CDC. Smoking prevalence and cessation in selected states, 1981-1983 and 1985-the behavioral risk factor surveys. MMWR 1986; 35: 74&743. Connor H, Marks V. Alcohol and diabetes: A position prepared by the nutrition subcommittee of the British Diabetic Association’s Medical Advisory Committee and approved by the Executive Council of the British Diabetic Association. Hum Nutr 1985; 39A: 393-399. Harris M, Entmacher P. Mortality from diabetes. In Diabetes in America: Diabetes Data Compiled 1984. National Institutes of Health; 1985: (NIH publication no. 85-1468), 29: 148.1. Fuller JH, Elford J, Goldblatt, Adelstein AM. Diabetes mortality: a new light on an underestimated health problem. Diabetologia 1983; 24: 336341.
Vol. 41, No.
IO, pp. 999-1006,
0895-4356/88$3.00+ 0.00 Pergamon Press plc
1988
Printed in Great Britain
SENTINEL
DEATHS
HEALTH EVENTS SURVEILLANCE IN DIABETES AMONG
PERSONS UNDER DIABETES
AGE 45 WITH
DIANE BILD I,* LINDA S. GEE,’ STEVENM. TEUTXH,’ BARBARAGABELLA,~ BARBARAA. HUDSPETH,~RALPH M. SCHUBERT,~CHARLESW. GOLLMAR,’ DOUGLASJ. KAPLAN,~FREDERICKA. CONNELL’and JULIE C. WILL’ ‘Division of Diabetes Control and 2Epidemiology Program Office, Centers for Disease Control, Atlanta, Georgia, 3Colorado Department of Health, 4111inoisDepartment of Health, 5Kentucky Department of Health Services, 6Louisiana Department of Health and Human Resources and ‘Washington Department of Social and Health Services, U.S.A. (Received in revised form 28 March 1988)
Abstract-The pilot study for a sentinel health events surveillance system for deaths among persons under age 45 with diabetes was conducted in six States in 1984 and 1985. Two hundred and thirty-three events were identified. Information from death certificates, physicians, and families revealed that 22% died from acute complications of diabetes and 53% from chronic complications. Blood pressure measurement and urinalysis testing had been performed in the last year for almost all of the decedents, but other preventive practices were reported less frequently. Hypertension was present in 57% and of those, was not controlled in 73%. Forty-four percent were cigarette smokers at the time of death. Agreement between physicians and families was generally higher for clinical conditions than for care practices. This surveillance system appears to yield information about the health care of persons with diabetes not readily available from other sources, although modifications may be necessary before implementation. Diabetes cations
Chronic disease
Surveillance
Mortality
INTRODUCTION
Surveillance is a mechanism to systematically record and track disease in defined populations over time. Its purposes are to measure disease frequency, risk factors, and trends in diseases and risk factors, to generate hypotheses regarding causation and prevention strategies, and to form the basis for population-based control programs. The classic model disease surveillance is employed for infectious diseases, including the reporting and enumeration of cases with
*Reprint requests should be addressed to: Diane Bild, Division of Diabetes Control, Centers for Disease Control, 1600 Clifton Rd NE, Atlanta, GA 30333, U.S.A.
Health care
Diabetes compli-
analysis and dissemination of results on a routine basis. Chronic disease surveillance is more difficult to conduct because the incidence of chronic diseases is difficult to measure and is only one of many endpoints of interest. Disease onset is insidious, there are multiple disease endpoints for a single disease, and diagnostic criteria for disease endpoints are non-uniform and relatively elaborate. Death, perhaps the most commonly recorded chronic disease endpoint, is often a poor reflection of morbidity. At other end of the disease spectrum, the measurement of risk factors for chronic diseases is desirable, because risk factors relate to disease and can be targeted directly for intervention. However, chronic diseases have multiple risk factors and long latency periods between the
999
DIANE BILD et al.
1000
risk factor and disease onset. Despite the heavy toll in health care costs and human suffering from chronic diseases in the United States, there are few routine reporting mechanisms for them. Diabetes was the seventh leading cause of mortality in the United States in 1984 [1] and the twelfth leading cause of years of potential life lost [2]. Sources of data on diabetes include death certificates, periodic population-based surveys, hospital discharge summaries, and clinic-based studies or registries. No single, currently available data source has all the desired features of a surveillance system-timeliness, a community base (as opposed to clinic base), specificity for diabetes, data related to morbidity and health care factors, and systematic feedback to practitioners in the community. While a variety of data sources are necessary for chronic disease control, the surveillance system for sentinel health events in diabetes described in this paper is one system with potential to address these requirements. A sentinel health event has been described by Rutstein et al. as the occurrence of unnecessary disease, unnecessary disability, or untimely death [3]. These events, designated because of their potential to be prevented, may serve as a reflection of the quality of health care in the community and as a warning signal that the quality of care may need to be improved. Under the assumption that mortality is most preventable in the younger population, we have chosen to focus on persons who died under age 45 with diabetes. A surveillance system was designed to (1) identify persons who died under age 45 with diabetes and (2) assess morbidity and health care practices in this group. This paper describes the pilot study for this surveillance system which was conducted in six States. METHODS
Six State health department-based Diabetes Control Programs (DCP) participated in the pilot study (Colorado, Illinois, Kentucky, Louisiana, Michigan, and Washington). Death certificates were collected for periods of 4-8 months from July 1984 through July 1985. It was not possible to assure that all months of the year were represented equally among States. An event was defined as any death of a person under age 45, with diabetes mentioned anywhere on the death certificate. Using standardized data collection instruments, DCP investigators gathered basic demographic data
from death certificates and, when feasible, conducted follow-back surveys of both physicians associated with the decedent and family members to obtain clinical and health care-related information. Death certificate information was collected as recorded on the certificate. Followback information was obtained by mail, by telephone, or through face-to-face interview. The form contained 16 questions. Questions were phrased as appropriate to the understanding of either medical personnel or laypersons. For example, “myocardial infarction” was used for medical personnel and “heart attack” was used for laypersons. The presence or history of clinical conditions and use of care practices in the last year of life were ascertained through a “yes”, “no”, or “unknown” response. Hypertension was defined as blood pressure greater than 140/90 mmHg. Legal blindness was defined as visual acuity of 20/200 or poorer. Nephropathy was defined as persistent proteinuria in the absence of infection. Other questions asked were type of medication for diabetes (insulin, oral agent, both, neither, or unknown), whether blood pressure in the last year was 2 140/90 mmHg, height, weight, smoking habits (never smoked, ex-smoker, smoked< 1 pack per day, smoked > 1 pack per day, or unknown), the level of compliance with pharmacologic and non-pharmacologic therapies (excellent, average, poor, or unknown), and the average number of times per year the decedent had seen a physician. Using all information available, the cause of death in each event was classified by two of the authors (D.B. and L.G.) into one of three categories based on the one condition most contributory to death. The categories included those attributed to (1) acute complications of diabetes (diabetic ketoacidosis or coma, infection, and hypoglycemia), (2) chronic complications (heart disease, end-stage renal disease, and cerebrovascular disease), and (3) other causes (diabetes only, not diabetes-related, and unknown). The mortality rate for each State was calculated by using the number of deaths (from death certificates) in the numerator and national diabetes prevalence data by region (National Health Interview Survey, NCHS, 1980) and the time period (fraction of a year) in the denominator. The expected rate among all States was calculated based on 1982 national death certificate data [4]. Chi-square analysis was per-
1001
Diabetes Sentinel Health Events Surveillance Table 1, Cause of death among 233 persons under 45 with diabetes, by age group Age in years Cause of death Acute complication Chronic complication Diabetes only, unrelated, or unknown Total @‘I
t&24 WI
25-34 (%I
3544 W)
Total W)
50.0
22.1
18.4
21.9
11.1
48.5
60.5
53.2
38.9
29.4
21.1
24.9
100.0 (18)
100.0 (68)
100.0 (147)
100.0 (233)
formed for comparison of categorical variables including demographic characteristics, clinical conditions, and health care practices. The Wilcoxon rank sum test was used to compare differences in distributions in variables which were not normally distributed. The degree of agreement between physician and family sources on survey data items was calculated using the kappa statistic, an index of agreement corrected for chance-expected agreement [5]. RESULTS
Two hundred and thirty-three events were identified in six States. Assuming the periods of data collection used by the States were representative of the entire year, the average observed annual mortality rate was 32.7 per 10,000 persons under age 45 with diabetes. The expected rate was 34.8 per 10,000 persons under age 45 with diabetes. The death certificate was the only source of information for 37% of the events. Follow-back information from physicians was available for 49% of the events and from families for 35% of the events. Both forms of follow-back information were available for 22% of the events. Decedents included 146 males and 87 females. The male to female mortality ratio was 1.78, compared to 1.93 in the general population under age 45 (based on 1984 mortality statistics for the United States). Twenty-two percent of the decedents died from acute complications of diabetes, 53% died from chronic complications, and 25% died from “diabetes only” or unrelated or unknown causes (Table 1). The distributions of cause of death were significantly different among the three different age groups (x2 = 17.9, p < 0.01). In the
youngest group (0-24yr) 50% of the deaths were due to acute complications, while in the oldest group (35-44yr) 18% died from acute complications. In the youngest age group 11% died from chronic complications, while in the oldest age group 60% died from chronic complications. There was no significant difference in the distributions of cause of death between males and females (Table 2). The median duration of diabetes for those who died from chronic complications was significantly longer than for those who died from acute complications (median duration 23 as compared to 14 yr, Wilcoxon rank sum test, p < 0.01). Several conditions were recorded with notable frequency on death certificates or among “other” conditions in the follow-back surveys. These included alcohol abuse (13 events), suicide (5 events), and cardiomyopathy or congestive heart failure in the absence of ischemic heart disease (7 events). Among the 13 events in which alcohol abuse was reported, 8 of the decedents were male; 5 of the 13 died from acute complications, 3 from chronic complications, and 5 from other causes or diabetes only. In the 5 events in which suicide was the cause of death, all of the decedents were male. The degree of agreement between sources of follow-back information (kappa statistic) was greatest for the presence of the clinical conditions dialysis or transplantation (1 .OO), nephropathy (0.74), retinopathy (0.73) and lower extremity amputation (0.73) [Table 3(a)]. It was also high for whether the decedent had had a blood pressure check in the last year (0.79). Agreement was poorer for survey items on laboratory testing, funduscopic examination, and self care practices. There was poorest agreement on the decedent’s compliance with nonpharmacologic and pharmacologic therapy (0.13 and 0.14, respectively) [Table 3(b)]. The kappa statistic may be considered to reflect agreement beyond that expected by chance as
Table 2. Cause of death among 233 persons under age 45 with diabetes, by sex Cause of death
Male W)
Female WI
Acute Chronic Unknown/diabetes Total WI
20.6 54.1 25.3 100.0 (146)
24.1 51.7 24. I 100.0 (87)
only
Table 3(a). Degree of agreement between physicians and families on clinical condition and care practices in the last year of life among persons with diabetes who died under age 45
Condition or practice Retinopathy Legal blindness Nephropathy Dialysis or transplant Lower extremity amputation Stroke Myocardial infarction History of hypertension Urinalysis in last year Glycohemoglobin test in last year Funduscopic examination in last year Blood pressure check Home glucose monitoring
Number of paired responses
Percent “yes” from family
Percent “yes” from physician
Agreement (kappa statistic)
43 41 41 43 41 38 43 44 36
67.4 36.6 61.0 37.2 17.1 10.5 30.2 52.3 94.4
69.8 39.0 68.3 37.2 14.6 5.3 18.6 65.9 94.4
0.73 0.64 0.74 1.00 0.73 0.64 0.60 0.63 0.46
18
77.8
38.9
0.31
2 39
58.3 95.0 69.2
80.6 92.5 53.8
0.26 0.79 0.26
There was a total of 50 events in which both sources of information were available. The number of paired responses excludes unknown and missing data. Table 3(b). Degree of agreement between physicians and families on clinical condition and care practices in the last year of life among persons with diabetes who died under age 45 Number of paired responses
Condition or practice Medication for diabetes
Family and physician responses Family
46
insulin oral both neither
Agreement (kappa statistic) 0.70
82.6 4.3 6.5 6.5
Physician
insulin oral both neither Cigarette smoking
84.8 6.5 4.3 4.3 0.52
40 Family
never ex-smoker current
57.5 22.5 20.0
Physician
never ex-smoker current Compliance-non-drug
therapy
60.0
10.0 30.0 0.13
36 Family
excellent average poor
33.3 44.4 22.2
Physician
excellent average poor Compliance-drug
therapy
16.7 38.9 44.4 0.14
37 Family
excellent average poor
70.3 24.3 5.4
Physician
excellent average poor
32.4 40.5 27.0
There was a total of 50 events in which both sources of information were available. The number of paired responses excludes unknown and missing data. 1002
1003
Diabetes Sentinel Health Events Surveillance Table 4. Prevalence of clinical conditions among persons with diabetes who died under age 45, by cause of death Cause of death Acute
Unknown/ diabetes only
Chronic
Total
Condition
N
% Positive responses
N
% Positive responses
N
% Positive responses
N
% Positive responses
Retinopathy Blindness Nephropathy Dialysis or transplant Amputation Stroke Myocardial infarction Hypertension
26 21 28
50.0 22.2 35.7
69 64 69
81.2 42.2 72.5
31 35 34
46.0 8.6 35.3
132 126 131
65.2 28.6 55.0
29 30 30
17.2 16.1 6.1
68 69 65
50.0 21.7 12.3
36 36 35
5.6 5.6 2.9
133 135 130
30.8 16.3 8.5
28 26
7.1 23.1
70 73
47.1 74.0
35 33
11.4 45.4
133 132
29.3 56.8
Prevalence of the various factors was calculated using physician responses first, and family responses if no physician response was available. N is the total number of responses excluding unknowns.
follows: values greater than 0.75 represent excellent agreement beyond chance, values below 0.40 represent poor agreement beyond chance, and values between 0.40 and 0.75 represent fair to good agreement beyond chance [S]. Families assessed decedents as having less nephropathy, more heart attacks (compared to myocardial infarctions, the term used for physicians), less high blood pressure (compared to hypertension, the term used for physicians), and using home glucose monitoring more than did physicians. They also tended to rate patients better than did physicians on compliance with both drug and non-drug therapies. Among family respondents the response concerning glycohemoglobin testing (described also as “blood test which determines blood sugar level over a long period of time”) included mostly unknowns. The prevalence of diabetic complications was generally high, as shown in Table 4. Among the most frequent complications were retinopathy (65%), nephropathy (55%), end-stage renal disease (31%) and hypertension (57%). The prevalence was higher among those who died from chronic complications than among those who died from acute complications. Hypertension was not controlled (reported as greater than 140/90 mmHg in the last year) in 73% of those who were reported to have had hypertension (Table 5). Almost all of the decedents had had blood pressure measurement and a urinalysis in the last year of life (Table 6). Among the 87% of the decedents who used insulin, 44% had had a glycohemoglobin measured in the last year and 50% had used self-monitoring of blood glucose.
Almost 44% were smokers at the time of death. While 79% had had an ophthalmic exam in the last year, it had been performed by an ophthalmologist in only 48% of these cases. DISCUSSION
It has been estimated that, with currently available technology for diabetes care, between 50 and 85% of complications of diabetes are treatable and preventable [6]. For example, it has recently been shown that the rate of decline in renal function may be slowed approximately 60% by aggressive antihypertensive management in selected patients with diabetes [7,8]. Previous studies on premature mortality among persons with diabetes have found a significant proportion of the deaths were due to preventable factors. In Washington State, the mor-
Table 5. Hypertension history and treatment, blood pressure measurement, and hypertension control among persons with diabetes who died under age 45
Hypertension history Hypertension history and under treatment Hypertension history and blood pressure checked in last year of life Hypertension history and blood pressure > 140/90 mmHg
N
% Positive responses
132
56.8
54
90.7
56
100.0
41
13.2
Prevalence of each condition was calculated using physician responses first, and family responses if no physician response was available. N is the total number of responses excluding unknowns.
1004
DIANE BILD et al.
Table 6. Care practices by sex in the last year of life among persons with diabetes who died under age 45 Female
Male
Blood pressure check Urinalysis Current smoker Glycohemoglobin test among insulin users Self-monitoring of blood glucose among insulin users Funduscooic examination Funduscopic examination performed by an ophthalmologist Median number of physician visits per year
Total
N
% Positive responses
N
% Positive responses
N
% Positive responses
61 61 75
91.8 86.9 44.0
38 37 42
100.0 100.0 42.9
99 98 117
95.0 91.8 43.6
53
39.6
26
53.8
79
44.3
64 63
48.4 76.2
38 36
52.6 83.3
102 99
50.0 78.8
28 N 79
39.2 Visits
18 -N
61.1 Visits
46 N
47.8 Visits
8
45
7
124
8
Prevalence of the various factors was calculated using physician responses first, and family responses if no physician response was available. N is the total number of responses excluding unknowns.
tality rate for persons under age 45 with diabetes has been reported to be eight times that of the general population under age 45 [9]. A detailed review of death certificates among persons under age 45 with diabetes found almost one-third of the deaths were due to acute complications for which there is definitive therapy. In a study in Great Britain, “neglect of diabetes” contributed to 27% of the deaths among 447 persons with diabetes who died under age 50 [lo]. Thus, it appears that improvements in health care may have an impact on the population of persons with diabetes under age 45. An examination of the morbidity and health care of the segment of the group that died may provide clues to how health care could be improved. Rates of complications of diabetes among the decedents in this study were very high. Over half had diabetic nephropathy, a condition associated with a 3 to 4-fold excess mortality rate [ 111. Clinical intervention (e.g. improved glycemic control and hypertension control) are necessary at least several years prior to the time of death to alter the course of this disease. Younger decedents were at greater risk from acute complications than older decedents. This may be due to a combination of death due to ketoacidosis at diagnosis, poor self-management of insulin, and the fact that chronic complications take more time to develop. Rates of such routine care practices as measurement of blood pressure and urinalysis testing were high, although one would expect to find that all persons with diabetes had had such basic exams, especially during the last year of
life. At this time health care needs were great because of clinically manifest complications of diabetes. However, while blood pressure was being measured routinely, hypertension was not controlled in a substantial proportion of these persons. The rates of glycohemoglobin testing and self blood glucose monitoring among those on insulin were relatively low (44 and 50%, respectively). This may be a reflection of suboptimal quality of care, as one might expect that virtually all persons on insulin use these technologies, especially those under 45. The prevalence of smoking of 44% was higher than the prevalance of 28% for adult males and 25% for adult females in the general population reported in 1985 [12]. The combination of alcohol and diabetes appears to be a dangerous combination. Alcohol has well-observed adverse effects on the health of persons with diabetes, including metabolic disturbances and compromise of the discipline and self-care integral to overall diabetes care [13]. A modest seasonal trend in diabetes mortality has been described with higher rates occurring during the winter months [4]. This may have had a minor effect on the calculated State mortality rates in this study, since States only monitored deaths for part of the year. Ideally, a surveillance system which seeks information on morbidity would obtain information from sources of health care and/or recipients of health care-hospitals, outpatient care centers, and community surveys. Unfortunately, data from these sources are not uniformly available, systematized, linked, or
1005
Diabetes Sentinel Health Events Surveillance
timely. As demonstrated in this pilot project, the death certificate provides a uniformly available, standardized means of identifying persons with diabetes, may be used to obtain data on mortality rates and trends directly, and can serve as an entry point to retrospectively obtain morbidity data. Death among persons under age 45 with diabetes is just one possible event which may be used as an indicator of the overall quality of diabetes care. Diabetic ketoacidosis and hypoglycemia are considered highly treatable with available technology and preventable with adequate monitoring and education; therefore deaths due to acute complications have been suggested as such a measure [3,9]. Other specific causes of death might also be monitored, such as death due to end-stage renal disease. In this pilot study the younger segment of the diabetic population was targeted to increase the proportion of preventable deaths represented among the events, because the risk of the complications of diabetes and mortality increase with age. However, it would be expected that remediable problems discovered in the health care for the diabetic population under age 45 also exist in the rest of the diabetic population. While in the present study each death was considered a sentinel health event, this is probably only appropriate for conditions for which the expected mortality rate is zero. For diabetes surveillance, death rates may be more appropriate units of measurement. Rates may be monitored, and if an unusually high rate is detected, data on decreased individuals may then be collected. It would be desirable to have established expected rates of mortality and normative standards for complications and care practices before launching follow-back studies. Surveillance information must be communicated back to local communities and used to further investigate possible problems, e.g. conduct case-control studies or more intensive monitoring of a suspected problem in a defined region. New systems for surveillance must be evaluated for the sensitivity of the system, the quality of data generated, the potential uses of the system, and the feasibility of implementation. The actual sensitivity of the system for identifying events is probably not as important as that it remains relatively constant. The sensitivity of this system in identifying appropriate events will depend on the completeness of recording of diabetes of death certificates and ascertainment
of death certificates, and on the reliability and completeness of follow-back survey information. Diabetes is listed on death certificates for only about half of the deaths among persons with diabetes [14], and this may vary by State and over time. Diabetes is more often listed as the underlying cause of death in younger persons than older persons with diabetes [15]. It may be assumed that the probability of diabetes being listed anywhere is also greater at younger ages. Thus, the use of a relatively young group of persons in the diabetic population increases the sensitivity of the system. In a related project in Michigan conducted during and after this pilot study, 77 of 84 events (91.7%) identified from both death certificates and hospital records were identified from death certificates alone. In seven of the 84 events there was no mention of diabetes on the death certificate (M. Halpern, personal communication). This provides some evidence that death certificates provide a fairly high degree of sensitivity in identifying these events. The major potential limitation of this system is the labor-intensiveness-the complexity and the time involved (between 4 and 6 hr per completed investigation in one State). Identification of events was relatively easy, but the follow-back involved active information tracking and retrieval. In the implementation of this system some survey data will be deleted and several options for reducing the burden of investigation have been proposed. Options for the latter include conducting follow-back investigation only when an unusual rate of events has been identified, investigating only a subsample of events, limiting sources of followback information to either physicians or families, and limiting the definition of an event to more specific diagnoses, such as death due to diabetic ketoacidosis. Further experience in sentinel health event surveillance will allow this system to be a useful instrument of investigation and, ultimately, to improve the health care system. Acknowledgements-We wish to thank Martin Halpern, PhD. John Beaslev. MA. and Steven Lonaabauah. MS. Michigan Department of Public Health, who gathered and analyzed data in Michigan and assisted in the preparation of this manuscript.
REFERENCES 1.
National Center for Health Statistics. Advance report of final mortality statistics, 1984. Monthly Vital Statistics Report. Hyattsville, Md.: Public Health Service; 26
1006
2.
3.
4.
5. 6.
I.
8.
DIANE BILD er al.
September 1986: Vol. 35, No. 6, Suppl. 2. DHHS Pub No. (PHS) 861120. CDC. Premature mortality from diabetes-use of sentinel health event surveillance to assess causes. MMWR 1986; 35: 711-714. Rutstein DD, Berenberg W, Chalmers TC et al. Measuring the quality of medical care: a clinical method. N Enel J Med 1976: 294: 582-588. National Cent& for Health Statistics. Advance report of final mortality statistics, 1982. Monthly Vital Statistics Report. Hyattsville, MD.: Public Health Service; 20 December 1984; Vol. 33, No. 9, Suppl. 2. DHHS Pub. No. (PHS) 85-1120. Fleiss JL. Statistical Methods for Rates and Proportions. New York: John Wiley; 1981: 2nd edn, 221-225. Carter Center. Closing the gap: The problem of diabetes in the United States. Diabetes Care 1985; 8: 391406. Mogensen CE. Long-term antihypertensive treatment inhibiting progression of diabetic nephropathy. Br Med J 1982; 285: 685688. Parving HH, Andersen AR, Smidt UM, Svendsen PA. Early aggressive antihypertensive treatment in kidney function in diabetic nephropathy. Lancet 1983; 1: 1175-l 178.
9.
10.
Il.
12.
13.
14.
15.
Connell FA. Epidemiologic approaches to the identification of problems in diabetes care. Diabetes Care 1985; 8(Suppl. 1): 82-86. Tunbridge WMG. Occasional Survey: Factors contributing to deaths of diabetics under fifty years of age. Lancet 1981: 569-572. Deckert T, Paulsen JE, Larsen M. Prognosis of diabetics with diabetes onset before the age of thirtyone. 1. survival, causes of death, and complications. Diabetologia 1978; 14: 363-370. CDC. Smoking prevalence and cessation in selected states, 1981-1983 and 1985-the behavioral risk factor surveys. MMWR 1986; 35: 74&743. Connor H, Marks V. Alcohol and diabetes: A position prepared by the nutrition subcommittee of the British Diabetic Association’s Medical Advisory Committee and approved by the Executive Council of the British Diabetic Association. Hum Nutr 1985; 39A: 393-399. Harris M, Entmacher P. Mortality from diabetes. In Diabetes in America: Diabetes Data Compiled 1984. National Institutes of Health; 1985: (NIH publication no. 85-1468), 29: 148.1. Fuller JH, Elford J, Goldblatt, Adelstein AM. Diabetes mortality: a new light on an underestimated health problem. Diabetologia 1983; 24: 336341.