Diabetes and influenza – epidemiology – OUTDATED – SEE PROPOSAL_BACKGROUND.doc

Watkins, 1970

-  Case series of 29 patients with ketoacidosis, concomitant in time with an influenza epidemic from 1969-1970. Influenza cannot be proved to be the precipitating cause, but circumstantial evidence was “considerable.” Earlier reference to FitzGerald et al., 1961, who suggested that 9% of all keto-acidosis cases were the result of respiratory infections – it is suggested by this study that influenza may disturb diabetic control more than others.

-  3/19 patients may have been type 2 (“non-ketotic diabetes, two being finally controlled with oral hypoglycaemics”) – none of these patients died.

-  7 patients died, although only 3 of these deaths occurred in the acute phase of acidosis (death due to hypokalemia) – the other causes of death were cardio/cerebrovascular, pulmonary emboli, and meningitis.

Bouter, 1991

-  Hospital-based cohort study with duodenal ulcer patients as controls.

-  Relative risks compared between epidemic and non-epidemic years, with non-epidemic years acting as a negative control.

-  Absolute rates compared between epidemic and non-epidemic years to attain a sense of absolute risk, with non-epidemic years standing in for the counterfactual.

-  Outcomes = risk of hospitalization with influenza, pneumonia, risk of mortality during hospitalization. RRs adjusted for age and sex, although method of doing this is not reported. No stat. testing for most RRs due to “census”.

-  RR hosp. influenza = 1.1 and 1.0 vs 5.7 and 6.2.

-  RR hosp. pneumonia = 20.3 and 15.8 vs 25.6 and 25.6.

-  RR hosp. leading to death = 42.4 and 91.8 vs 30.9 and 31.8.

-  RR hospitalization for diabetic keto-acidosis, epidemic (1978) vs non-epidemic (1976) year = 15.9. Using 1977 as the control year RR = 13.2.

-  Risk of death among those hospitalized for pneumonia or DKA higher in epidemic vs non=epidemic years. RR death pneumonia = 14.6% vs 25.7% (p < 0.01). RR for death DKA = 14.7% vs 25.4% (p < 0.01). However, few patients died of influenza. Because pneumonia hospitalizations increased during influenza epidemic, it is likely that influenza cases are getting diagnosed as pneumonia.

-  Study supports both increased incidence of ARI in diabetes, as well as increased mortality after ARI in diabetes.

-  First study to demonstrate increased mortality (?).

Bisno, 1971

-  Cohort of CAP patients during the 1968-1969 Influenza A(2) Hong Kong epidemic, admitted to hospital in Memphis. This included 98 medicine admissions.

-  83% of patients with confirmed LRTI during this period had serologic signs of influenza infection.

-  Study mentions that ~three forths of patients with influenza-associated pneumonia had identifiable underlying medical disorders or were pregnant. DM was listed as the third most common category, 8 patients out of 88.

-  A control year was selected – 1969-1970, no influenza epidemic. Only 29 medicine admissions over the same period. Underlying chronic illness, including diabetes mellitus, noted in 23 (80%) of patients. DM may be a RF for CAP admission in patients with or without influenza.

-  Case fatality during the epidemic was 14/106 (13%).

Stocks, 1935

-  Death certificate analysis, 1921-1933.

-  Months of first quarters in each year (3 per year), divided into months with normal, mild-moderate, or severe levels of influenza epidemic (by number of influenza deaths).

-  Table 2: Stratify on month, look at the trend in other disease deaths with increasing influenza epidemic severity (ordinal ranking of years for each month). Other disease deaths expressed as a percentage increase over the expected disease deaths for that period. Expected disease deaths estimated by linear regression for a secular trend pinned by two points at either end of the period in question. (Adjusted for secular trends).

-  Fig. 2: Assess whether cold vs warmer weather is a confounder of the relation in table 2 by stratifying on cold vs warm, and graphing other disease deaths (percent of expected due to secular trends) vs increasing influenza severity.

-  Table 3: Simple analysis, mean number of deaths from other diseases by increasing severity of influenza epidemic.

-  Table 4: Same as in table 3, but analyzed by months bucketed into normal, mild/moderate, and severe epidemic categories. Influenza-attributable fraction of deaths calculated by subtracting mild/moderate and severe from normal numbers.

-  Table 5: To assess adequacy of death certificate reporting.

-  Table 6: Similar point as table 4, but subtracting peri-epidemic from epidemic quarters in each year, and then obtaining the difference for years with an epidemic vs years of low influenza. Extrapolation to 1933 to assess fit.

-  Many disease show rises in deaths contemporaneous with the presence and increasing severity of influenza activity, even in analyses stratified for warm/cold, month of year, and with adjustment for secular trends.

-  Diabetes was not one of them. For diabetes and certain other chronic diseases, death would be assigned to influenza, primarily, but assigned to the underlying disease if influenza had not been recognized. These influences have contrary effects on the behavior of death certificate frequencies, assuming additional or hastened deaths due to influenza, so these results cannot be interpreted.

Carrat, 1995

-  Objectives – estimate influenza-associated mortality for a variety of death certificate causes of death, and estimate rate of deaths avoided by vaccination.

-  Death certificate data, analysis proceeded by modified ARIMA to account for serial auto-correlation. Linear component of deaths regressed against influenza activity, with a different regression equation estimated for each year. Unsure how seasonality is corrected for in this model. Deaths from 1980-1990. Associated component was small or non-significant for non-epidemic years – good negative control. Test of fit by average residuals good.

-  Rate of deaths avoided estimated using a formula and a sensitivity table.

-  The number of deaths attributable to influenza ranged from 0.5 to 8.2 times the number of deaths registered as due to influenza alone. Attributable deaths especially important for respiratory diseases, cardiovascular disease, and chronic renal failure. Less so for diabetes and lung cancer.

-  Diabetes: Significantly associated influenza component in only 2 of 5 epidemic years, associated rate = 6.2/100 000 and 5.6/100 000 compared with 14.7 to 29.8 for ischemic heart disease and 16.1 to 40.7 for pneumonia.

-  However, death due to diabetes may have been restricted to keto-acidosis in type 1 patients, which is less common in any event. Authors include diabetes in the general listing of “high risk conditions” with increased risk of death from influenza. NOTE however that denominator = general population, and that there is no valid comparison to demonstrate an increased risk of death in high risk patients – compared to who?

Giles, 1957

-  Necropsy results for a case series of 46 patients admitted for pneumonia, in which influenza was either the primary or a contributing cause of death. Patients were selected from 53 deaths at City General Hospital in Stoke-on-Trent during an epidemic with a high attack rate in September-October, 1957.

-  2/46 deaths occurred in patients with diabetes (4.3%).

-  The fatality rate of patients admitted to hospital during the epidemic exceeded 25% despite intensive antibiotic therapy.

Stuart-Haris, 1950

-  Case series of 85 influenza-linked deaths during an epidemic period, and 11 influenza-linked deaths during an inter-epidemic period. Cases and controls were similar by age and sex. Also examination of 22 fatal illnesses in Sheffield during the 1949 influenza epidemic.

-  On death certification / medical records recounting the acute event, influenza correlated highly with other respiratory disease and with cardiovascular diseases. It appeared random as to whether influenza was recorded as contributing or primary cause of death, relative to the underlying condition. E.g.: Pneumonia mentioned in 40 instances, heart failure in 60 instances.

-  Sheffield cases: 9/22 fatal illnesses obtained positive viral cultures. 1/9 viral deaths had diabetes recorded. 1/13 virus-negative deaths had diabetic coma recorded. OR = 1.50, p = 1.00.

Collins, 1930

-  Objectives: study the course of recorded mortality from pneumonia and influenza, identify excess mortality, measure the excess, study its distribution across geographic regions of the US, and to study the movement of epidemics from one region to another.

-  Weekly pneumonia and influenza death certifications from 95 US cities of >100 000 population taken to be geographically representative. Excess mortality estimated by removing expected seasonal mortality. Expected seasonal mortality estimated by 5 period moving average model of the medians for each period, including epidemic and non-epidemic periods (epidemic peaks included, but tamped down by non-epidemic values and by the moving average function). Total excess epidemic deaths / 100 000 measured. Length of epidemic measured by IQR for deaths.

-  Time period – 1920-1929, in which 6 epidemics occurred. Excess mortality = 250 000 total.

-  Note the collocation in time of pneumonia and influenza peaks.

-  Collins also known for pioneering the use of excess mortality to define the start and end-points of an influenza epidemic.

Collins, 1932

-  Author noted that studies of influenza mortality examine deaths credited to influenza or pneumonia – clinical grounds for specificity assumed, screen out the effects of coincidental epidemics of other diseases. However, some evidence had emerged that excess deaths from all other causes were also increased during influenza epidemics.

-  Objective – Measure the excess for all other causes in addition to pneumonia and influenza. Provide evidence that the excess for all other causes is due to influenza infection, and not coincidental epidemics. Examine the contributions of other causes of death to excess epidemic mortality.

-  Data – weekly death data from 35 cities from 1917-1929.

-  Methods as in Collins, 1930. Seasonal trends for particular non-PI causes of death estimated by averaging values for the same period of time from the previous and following years (very back of hand calculations).

-  Excess mortality due to non-PI causes substantial, shown to be quite contemporaneous with PI excesses. Time properties (peak day, 25th, 50th, 75th percentile of deaths, IQR) quite similar.

-  Percentage of total excess mortality credited to causes other than PI – about 40% for all epidemics other than 1918-19 (8%) and 1920 (23%) (much lower percentage in those major epidemics).

-  Death certificate analysis – less than half of the excess deaths credited primarily to causes other than PI listed PI as a contributory cause.

-  Similar analysis for particular non-PI causes of death. Peaks in total deaths for diabetes corresponded in time to influenza peaks, but were smaller and less definite than in the case of organic heart disease.

-  During epidemics of 1922, 1923, 1926, 1928, and 1928-29, diabetes accounted for 6.3% of excess mortality from non-PI causes, and 6.5/100 000 excess mortality.

-  Coincidence of diabetes mortality and influenza suggests an association. Diabetes listed among the “chief causes” of excess deaths other than PI – may be the origins of the “high risk group” designation here. However, it cannot be concluded that diabetes is a high-risk group because there lacks a valid low-risk comparison. Additionally, diabetes deaths likely to be keto-acidosis deaths in type 1 DM. Absolute contribution to numbers of deaths low.

Serfling, 1963

-  Methods paper introduces methods for quantifying excess mortality and the epidemic threshold. Key development: use the method of partitioning mortality for prospective, in addition to retrospective, analysis.

-  Age of low access to computing power.

-  Mortality best indicator of influenza activity, despite 4 week lag behind increase in morbidity. Reporting through surveillance channels, laboratory diagnostic tests, industrial and school absenteeism not idea. Nowadays, these reasons may not hold.

-  Excess is the additional amount after subtracting seasonal and secular trends.

-  Secular trends may be estimated by simple linear regression.

-  Seasonality requires Fourier terms. A single Fourier term is sufficient.

-  Simple least squares regression on non-epidemic years to extrapolate expected mortality in epidemic years.

-  Y = b1 + a1*t + a2*cos((2(pi)*t/(N) – b2)

-  Suggested – a two stage regression, since doing it all in one go restricts the estimation to non-epidemic years only, whereas the secular trend may be estimated from non-epidemic months in epidemic years as well. Reduces extrapolation.

-  Steps: Estimate secular trend. Remove secular trend from the data. Estimate seasonal change. Restore secular trend component.

-  A method of performing this involving only simple addition and one division is provided – method of double integration.

-  Need to distinguish epidemic increase from regular random variation. Standard deviation proposed – 2 weeks with 1.65xSD over the expected number of deaths.

-  Assuming null hypothesis – no epidemics – the chances of one or more instances in which the epidemic threshold is exceeded is 0.75. The chance of two consecutive excesses over 26 weeks is approx. 0.10.

-  For a moderate sized epidemic, the risk of failing to detect is 0.9 at the end of the second week, but only 0.1 at the end of the third.

Housworth and Langmuir, 1974

-  Analysis of excess mortality, building on the tradition of Farr, Frost, Pearl, Collins, and Serfling.

-  1957-1966 monthly deaths.

-  Expected mortality rates estimated as in Serfling, except with numerous Fourier terms and an additional quadratic term for the secular trend. Least-squares fit.

-  Innovation – Relative intensity statistics. Additive standard deviations of the expected deaths overall and between mutually exclusive causes of death devised by regressing on the same non-epidemic time periods for all causes. Relative intensity is the sum of standardized residuals over the length of the epidemic, and, under the hypothesis of no epidemic, is a normal variable with zero mean and variance equal to the number of contiguous time periods corresponding to an epidemic. Relative intensities allow comparisons of epidemic severity between epidemics, and severity of mortality between causes of death within epidemics. Comparisons can now be tested statistically. Major limitation: The expected mortality for the entire multi-epidemic period must be fit with a single model.

-  Note that relative intensities may classify as significant a cause of death with fewer excess deaths than one classified as non-significant. Examine percent excess from expected to see why.

-  Confirmed observation by Collins that respiratory deaths account for merely 40% of excess mortality. The proportion varied from 30 to 38% in more severe epidemics, to less than 25% in milder epidemics. The majority of excess deaths are due to heart, circulatory, and nervous causes.