- Split View
-
Views
-
Cite
Cite
Lisa A Jackson, Jennifer C Nelson, Patti Benson, Kathleen M Neuzil, Robert J Reid, Bruce M Psaty, Susan R Heckbert, Eric B Larson, Noel S Weiss, Functional status is a confounder of the association of influenza vaccine and risk of all cause mortality in seniors, International Journal of Epidemiology, Volume 35, Issue 2, April 2006, Pages 345–352, https://doi.org/10.1093/ije/dyi275
- Share Icon Share
Abstract
Background Functional status limitations may be associated with both an increased risk of death and a decreased likelihood of influenza vaccination, and so may confound the association of influenza vaccination and risk of all cause mortality in seniors.
Methods We conducted a nested case–control study of persons ≥65 years of age that included 252 cases who died during an influenza season and 576 age-matched controls. We identified functional limitations by medical record review, and compared the effect of adjustment for those factors with that of adjustment for disease covariates defined by diagnosis codes, using methods reported by previous influenza vaccine effectiveness studies, on the association of influenza vaccination and risk of death.
Results Functional limitations, such as requiring assistance for bathing, were highly prevalent in cases, even in the subgroup defined as free of comorbidity by diagnosis code criteria, and were associated with a decreased likelihood of vaccination among controls. Adjustment for functional limitations resulted in an estimate of the relative risk of death in vaccinated persons compared with unvaccinated persons that was closer to the null [odds ratio (OR), 0.71; 95% confidence interval (95% CI), 0.47–1.06] than the unadjusted estimate (OR, 0.59; 95% CI, 0.41–0.83). In contrast, adjustment for diagnosis code covariates moved the estimate further from the null (OR, 0.45; 95% CI, 0.30–0.68).
Conclusions Functional limitations appear to be important confounders of the association of vaccination and risk of death, while adjustment for diagnosis code covariates did not control for a healthy vaccinee bias. Further research is needed on methods to reduce the influence of bias in observational studies of influenza vaccine effectiveness.
Observational studies have reported that seniors who receive influenza vaccine are at substantially lower risk of all cause mortality during influenza season than unvaccinated seniors.1–19 It is possible that differences in health status between seniors who were vaccinated and those who were not could account for some or all of the observed reduction in risk. In a retrospective cohort study of over 70 000 seniors enrolled in Group Health Cooperative between 1995 and 2003, we found that influenza vaccination was associated with a 44% [95% confidence interval (CI), 39–48%] reduction in risk of death during influenza season, which is consistent with previously reported estimates. In contrast to other cohort studies, however, we also evaluated the association during the pre-influenza period, and found that vaccination was associated with a 61% (95% CI, 53–67%) reduction in risk of death before the influenza season. The reduction in risk in the pre-influenza period indicates the presence of bias due to preferential receipt of the vaccine by relatively healthy seniors.
In the pre-influenza period, complete adjustment for this healthy vaccinee bias should lead to an estimate of the relative risk of death in vaccinated persons compared with unvaccinated persons that is close to the null value of 1.0. During influenza season, adjustment for this bias should lead to an estimate of the relative risk that is closer to the null than the unadjusted estimate. In the cohort study, we evaluated the effect of adjustment for covariates defined primarily by groupings of diagnosis codes assigned to medical encounters, such as those for cancer or heart disease, according to methods described in previous influenza vaccine effectiveness studies.1–12 We found that this method of adjustment resulted in estimates of the relative risk of death in vaccinated persons compared with unvaccinated persons that were further from the null (towards a greater vaccine effect) than the unadjusted estimates, in both the pre-influenza period as well as in the influenza season. Our findings therefore indicate that adjustment for these diagnosis code covariates does not properly adjust for confounding in the association of influenza vaccine and risk of all cause mortality.
We hypothesized that functional status limitations, such as requiring assistance to ambulate or bathe, may be associated with an increased risk of death20–23 and with a decreased likelihood of influenza vaccination in seniors.15 To explore the hypothesis that functional status limitations are confounders of the association of influenza vaccination and risk of death, we conducted a nested case–control study among the cohort of Group Health seniors, comparing 252 cases who died during the 1997/1998 influenza season with 576 age-matched controls. We identified functional status limitations by chart review, and compared with effect of adjustment for those factors with the effect of adjustment for covariates defined by groupings of diagnosis codes using previous methods, as well as with the effect of adjustment for diagnosis code covariates designed to be more specific indicators of disease severity.
Methods
Study population
Cases and controls were identified from among the 36 128 members of Group Health Cooperative, a large health maintenance organization (HMO) in Washington State, who were ≥65 years of age as of January 1, 1998, had been continuously enrolled since at least January 1, 1993, had at least one record of service during 1997, and were not residents of a nursing home in 1997. Cases were defined as all persons who died in January through March 1998, which includes the period of influenza viral circulation for that season. For each case, two to three controls matched by age (year of age until age 89 years and then by age-groups 90–94, 95–99, and ≥100 years) and sex were randomly selected from the remaining members of the study population. Each control was assigned an index date, the date of death of their matched case.
The predominant influenza viral strain circulating in the 1997/1998 influenza season was not included in that year's trivalent inactivated influenza vaccine. Since the 1997/1998 influenza season was a year in which vaccine effectiveness may have been relatively low, we selected this year for our analysis to better assess the effect of confounding.
Covariate definitions
Covariates defined by groupings of diagnosis codes based on previous methods (Group 1 variables). Disease covariates, listed in Table 1, were defined by groupings of International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes assigned to inpatient and outpatient medical encounters in 1997, based on methods reported by published studies of influenza vaccine effectiveness in other HMO populations.3,5
. | . | . | . | Risk of death during January through March, 1998 in univariate matched case–control analysis . | . | . | . | Likelihood of vaccination among controls, adjusted for age and sex . | . | ||
---|---|---|---|---|---|---|---|---|---|---|---|
Category . | Variable . | % cases (n = 252) . | % controls (n = 576) . | OR . | 95% CI . | % vaccinated controls (n = 459) . | % unvaccinated controls (n = 117) . | OR . | 95% CI . | ||
Group 1: Variables defined by diagnosis codes using previously published methods | Atrial fibrillation | 16 | 7 | 2.41 | 1.51–3.84 | 8 | 6 | 1.42 | 0.60–3.36 | ||
Cancer | 32 | 15 | 2.83 | 1.95–4.10 | 17 | 7 | 2.53 | 1.16–5.48 | |||
Diabetes mellitus | 20 | 12 | 1.77 | 1.18–2.65 | 12 | 13 | 0.82 | 0.44–1.54 | |||
Dementia | 17 | 9 | 2.03 | 1.28–3.21 | 9 | 11 | 0.81 | 0.41–1.61 | |||
Heart disease | 45 | 29 | 2.09 | 1.52–2.88 | 30 | 25 | 1.38 | 0.85–2.23 | |||
Hypertension | 32 | 26 | 1.35 | 0.97–1.88 | 27 | 22 | 1.30 | 0.79–2.13 | |||
Lipid disorder | 5 | 7 | 0.76 | 0.42–1.39 | 8 | 3 | 2.38 | 0.82–6.93 | |||
Lung disease | 49 | 29 | 2.39 | 1.75–3.28 | 29 | 27 | 1.07 | 0.69–1.65 | |||
Renal disease | 13 | 4 | 3.67 | 2.09–6.45 | 4 | 3 | 1.98 | 0.55–7.06 | |||
Rheumatologic disease | 4 | 2 | 2.17 | 0.93–5.03 | 2 | 2 | 0.96 | 0.20–4.55 | |||
Group 2: Categorical variables defined by further stratification of diagnosis codes | No cancer | 68 | 85 | Ref | 83 | 93 | Ref | ||||
Non-serious cancer | 10 | 12 | 1.08 | 0.65–1.81 | 14 | 4 | 3.34 | 1.29–8.63 | |||
Serious or metastatic cancer | 22 | 3 | 10.68 | 5.54–20.59 | 3 | 3 | 1.15 | 0.31–4.27 | |||
No renal disease | 87 | 96 | Ref | 96 | 97 | Ref | |||||
Renal disease, no chronic renal failure | 7 | 2 | 3.18 | 1.54–6.55 | 3 | 1 | 4.41 | 0.55–35.14 | |||
Chronic renal failure | 6 | 2 | 4.28 | 1.86–9.83 | 1 | 2 | 0.92 | 0.18–4.71 | |||
No heart disease | 55 | 71 | Ref | 69 | 75 | Ref | |||||
Heart disease, no CHF | 21 | 23 | 1.28 | 0.88–1.86 | 25 | 14 | 1.89 | 1.07–3.35 | |||
CHF | 24 | 6 | 6.36 | 3.74–10.83 | 5 | 10 | 0.58 | 0.27–1.26 | |||
No diabetes | 80 | 88 | Ref | 88 | 87 | Ref | |||||
Diabetes, no complication | 16 | 11 | 1.63 | 1.05–2.53 | 11 | 8 | 1.13 | 0.54–2.34 | |||
Diabetes complications | 4 | 2 | 2.56 | 1.06–6.19 | 1 | 4 | 0.21 | 0.06–0.78 | |||
No lung disease | 51 | 71 | Ref | 71 | 73 | Ref | |||||
Lung disease, no chronic lung disease | 9 | 15 | 0.86 | 0.51–1.43 | 15 | 14 | 1.09 | 0.60–2.00 | |||
Chronic lung disease | 40 | 14 | 4.23 | 2.88–6.22 | 14 | 14 | 1.06 | 0.58–1.95 | |||
Group 3: Functional status variables defined by chart review | Diagnosis of dementia | 21 | 9 | 2.81 | 1.79–4.40 | 8 | 15 | 0.51 | 0.27–0.96 | ||
Requires assistance for bathing | 41 | 8 | 13.43 | 7.63–22.65 | 6 | 14 | 0.48 | 0.24–0.98 | |||
Requires assistance for ambulation | 45 | 15 | 5.45 | 3.69–8.07 | 13 | 22 | 0.63 | 0.36–1.09 | |||
Lives in a non-home setting | 19 | 9 | 2.80 | 1.75–4.49 | 8 | 13 | 0.55 | 0.30–1.01 |
. | . | . | . | Risk of death during January through March, 1998 in univariate matched case–control analysis . | . | . | . | Likelihood of vaccination among controls, adjusted for age and sex . | . | ||
---|---|---|---|---|---|---|---|---|---|---|---|
Category . | Variable . | % cases (n = 252) . | % controls (n = 576) . | OR . | 95% CI . | % vaccinated controls (n = 459) . | % unvaccinated controls (n = 117) . | OR . | 95% CI . | ||
Group 1: Variables defined by diagnosis codes using previously published methods | Atrial fibrillation | 16 | 7 | 2.41 | 1.51–3.84 | 8 | 6 | 1.42 | 0.60–3.36 | ||
Cancer | 32 | 15 | 2.83 | 1.95–4.10 | 17 | 7 | 2.53 | 1.16–5.48 | |||
Diabetes mellitus | 20 | 12 | 1.77 | 1.18–2.65 | 12 | 13 | 0.82 | 0.44–1.54 | |||
Dementia | 17 | 9 | 2.03 | 1.28–3.21 | 9 | 11 | 0.81 | 0.41–1.61 | |||
Heart disease | 45 | 29 | 2.09 | 1.52–2.88 | 30 | 25 | 1.38 | 0.85–2.23 | |||
Hypertension | 32 | 26 | 1.35 | 0.97–1.88 | 27 | 22 | 1.30 | 0.79–2.13 | |||
Lipid disorder | 5 | 7 | 0.76 | 0.42–1.39 | 8 | 3 | 2.38 | 0.82–6.93 | |||
Lung disease | 49 | 29 | 2.39 | 1.75–3.28 | 29 | 27 | 1.07 | 0.69–1.65 | |||
Renal disease | 13 | 4 | 3.67 | 2.09–6.45 | 4 | 3 | 1.98 | 0.55–7.06 | |||
Rheumatologic disease | 4 | 2 | 2.17 | 0.93–5.03 | 2 | 2 | 0.96 | 0.20–4.55 | |||
Group 2: Categorical variables defined by further stratification of diagnosis codes | No cancer | 68 | 85 | Ref | 83 | 93 | Ref | ||||
Non-serious cancer | 10 | 12 | 1.08 | 0.65–1.81 | 14 | 4 | 3.34 | 1.29–8.63 | |||
Serious or metastatic cancer | 22 | 3 | 10.68 | 5.54–20.59 | 3 | 3 | 1.15 | 0.31–4.27 | |||
No renal disease | 87 | 96 | Ref | 96 | 97 | Ref | |||||
Renal disease, no chronic renal failure | 7 | 2 | 3.18 | 1.54–6.55 | 3 | 1 | 4.41 | 0.55–35.14 | |||
Chronic renal failure | 6 | 2 | 4.28 | 1.86–9.83 | 1 | 2 | 0.92 | 0.18–4.71 | |||
No heart disease | 55 | 71 | Ref | 69 | 75 | Ref | |||||
Heart disease, no CHF | 21 | 23 | 1.28 | 0.88–1.86 | 25 | 14 | 1.89 | 1.07–3.35 | |||
CHF | 24 | 6 | 6.36 | 3.74–10.83 | 5 | 10 | 0.58 | 0.27–1.26 | |||
No diabetes | 80 | 88 | Ref | 88 | 87 | Ref | |||||
Diabetes, no complication | 16 | 11 | 1.63 | 1.05–2.53 | 11 | 8 | 1.13 | 0.54–2.34 | |||
Diabetes complications | 4 | 2 | 2.56 | 1.06–6.19 | 1 | 4 | 0.21 | 0.06–0.78 | |||
No lung disease | 51 | 71 | Ref | 71 | 73 | Ref | |||||
Lung disease, no chronic lung disease | 9 | 15 | 0.86 | 0.51–1.43 | 15 | 14 | 1.09 | 0.60–2.00 | |||
Chronic lung disease | 40 | 14 | 4.23 | 2.88–6.22 | 14 | 14 | 1.06 | 0.58–1.95 | |||
Group 3: Functional status variables defined by chart review | Diagnosis of dementia | 21 | 9 | 2.81 | 1.79–4.40 | 8 | 15 | 0.51 | 0.27–0.96 | ||
Requires assistance for bathing | 41 | 8 | 13.43 | 7.63–22.65 | 6 | 14 | 0.48 | 0.24–0.98 | |||
Requires assistance for ambulation | 45 | 15 | 5.45 | 3.69–8.07 | 13 | 22 | 0.63 | 0.36–1.09 | |||
Lives in a non-home setting | 19 | 9 | 2.80 | 1.75–4.49 | 8 | 13 | 0.55 | 0.30–1.01 |
CI, confidence interval; CHF, congestive heart failure.
. | . | . | . | Risk of death during January through March, 1998 in univariate matched case–control analysis . | . | . | . | Likelihood of vaccination among controls, adjusted for age and sex . | . | ||
---|---|---|---|---|---|---|---|---|---|---|---|
Category . | Variable . | % cases (n = 252) . | % controls (n = 576) . | OR . | 95% CI . | % vaccinated controls (n = 459) . | % unvaccinated controls (n = 117) . | OR . | 95% CI . | ||
Group 1: Variables defined by diagnosis codes using previously published methods | Atrial fibrillation | 16 | 7 | 2.41 | 1.51–3.84 | 8 | 6 | 1.42 | 0.60–3.36 | ||
Cancer | 32 | 15 | 2.83 | 1.95–4.10 | 17 | 7 | 2.53 | 1.16–5.48 | |||
Diabetes mellitus | 20 | 12 | 1.77 | 1.18–2.65 | 12 | 13 | 0.82 | 0.44–1.54 | |||
Dementia | 17 | 9 | 2.03 | 1.28–3.21 | 9 | 11 | 0.81 | 0.41–1.61 | |||
Heart disease | 45 | 29 | 2.09 | 1.52–2.88 | 30 | 25 | 1.38 | 0.85–2.23 | |||
Hypertension | 32 | 26 | 1.35 | 0.97–1.88 | 27 | 22 | 1.30 | 0.79–2.13 | |||
Lipid disorder | 5 | 7 | 0.76 | 0.42–1.39 | 8 | 3 | 2.38 | 0.82–6.93 | |||
Lung disease | 49 | 29 | 2.39 | 1.75–3.28 | 29 | 27 | 1.07 | 0.69–1.65 | |||
Renal disease | 13 | 4 | 3.67 | 2.09–6.45 | 4 | 3 | 1.98 | 0.55–7.06 | |||
Rheumatologic disease | 4 | 2 | 2.17 | 0.93–5.03 | 2 | 2 | 0.96 | 0.20–4.55 | |||
Group 2: Categorical variables defined by further stratification of diagnosis codes | No cancer | 68 | 85 | Ref | 83 | 93 | Ref | ||||
Non-serious cancer | 10 | 12 | 1.08 | 0.65–1.81 | 14 | 4 | 3.34 | 1.29–8.63 | |||
Serious or metastatic cancer | 22 | 3 | 10.68 | 5.54–20.59 | 3 | 3 | 1.15 | 0.31–4.27 | |||
No renal disease | 87 | 96 | Ref | 96 | 97 | Ref | |||||
Renal disease, no chronic renal failure | 7 | 2 | 3.18 | 1.54–6.55 | 3 | 1 | 4.41 | 0.55–35.14 | |||
Chronic renal failure | 6 | 2 | 4.28 | 1.86–9.83 | 1 | 2 | 0.92 | 0.18–4.71 | |||
No heart disease | 55 | 71 | Ref | 69 | 75 | Ref | |||||
Heart disease, no CHF | 21 | 23 | 1.28 | 0.88–1.86 | 25 | 14 | 1.89 | 1.07–3.35 | |||
CHF | 24 | 6 | 6.36 | 3.74–10.83 | 5 | 10 | 0.58 | 0.27–1.26 | |||
No diabetes | 80 | 88 | Ref | 88 | 87 | Ref | |||||
Diabetes, no complication | 16 | 11 | 1.63 | 1.05–2.53 | 11 | 8 | 1.13 | 0.54–2.34 | |||
Diabetes complications | 4 | 2 | 2.56 | 1.06–6.19 | 1 | 4 | 0.21 | 0.06–0.78 | |||
No lung disease | 51 | 71 | Ref | 71 | 73 | Ref | |||||
Lung disease, no chronic lung disease | 9 | 15 | 0.86 | 0.51–1.43 | 15 | 14 | 1.09 | 0.60–2.00 | |||
Chronic lung disease | 40 | 14 | 4.23 | 2.88–6.22 | 14 | 14 | 1.06 | 0.58–1.95 | |||
Group 3: Functional status variables defined by chart review | Diagnosis of dementia | 21 | 9 | 2.81 | 1.79–4.40 | 8 | 15 | 0.51 | 0.27–0.96 | ||
Requires assistance for bathing | 41 | 8 | 13.43 | 7.63–22.65 | 6 | 14 | 0.48 | 0.24–0.98 | |||
Requires assistance for ambulation | 45 | 15 | 5.45 | 3.69–8.07 | 13 | 22 | 0.63 | 0.36–1.09 | |||
Lives in a non-home setting | 19 | 9 | 2.80 | 1.75–4.49 | 8 | 13 | 0.55 | 0.30–1.01 |
. | . | . | . | Risk of death during January through March, 1998 in univariate matched case–control analysis . | . | . | . | Likelihood of vaccination among controls, adjusted for age and sex . | . | ||
---|---|---|---|---|---|---|---|---|---|---|---|
Category . | Variable . | % cases (n = 252) . | % controls (n = 576) . | OR . | 95% CI . | % vaccinated controls (n = 459) . | % unvaccinated controls (n = 117) . | OR . | 95% CI . | ||
Group 1: Variables defined by diagnosis codes using previously published methods | Atrial fibrillation | 16 | 7 | 2.41 | 1.51–3.84 | 8 | 6 | 1.42 | 0.60–3.36 | ||
Cancer | 32 | 15 | 2.83 | 1.95–4.10 | 17 | 7 | 2.53 | 1.16–5.48 | |||
Diabetes mellitus | 20 | 12 | 1.77 | 1.18–2.65 | 12 | 13 | 0.82 | 0.44–1.54 | |||
Dementia | 17 | 9 | 2.03 | 1.28–3.21 | 9 | 11 | 0.81 | 0.41–1.61 | |||
Heart disease | 45 | 29 | 2.09 | 1.52–2.88 | 30 | 25 | 1.38 | 0.85–2.23 | |||
Hypertension | 32 | 26 | 1.35 | 0.97–1.88 | 27 | 22 | 1.30 | 0.79–2.13 | |||
Lipid disorder | 5 | 7 | 0.76 | 0.42–1.39 | 8 | 3 | 2.38 | 0.82–6.93 | |||
Lung disease | 49 | 29 | 2.39 | 1.75–3.28 | 29 | 27 | 1.07 | 0.69–1.65 | |||
Renal disease | 13 | 4 | 3.67 | 2.09–6.45 | 4 | 3 | 1.98 | 0.55–7.06 | |||
Rheumatologic disease | 4 | 2 | 2.17 | 0.93–5.03 | 2 | 2 | 0.96 | 0.20–4.55 | |||
Group 2: Categorical variables defined by further stratification of diagnosis codes | No cancer | 68 | 85 | Ref | 83 | 93 | Ref | ||||
Non-serious cancer | 10 | 12 | 1.08 | 0.65–1.81 | 14 | 4 | 3.34 | 1.29–8.63 | |||
Serious or metastatic cancer | 22 | 3 | 10.68 | 5.54–20.59 | 3 | 3 | 1.15 | 0.31–4.27 | |||
No renal disease | 87 | 96 | Ref | 96 | 97 | Ref | |||||
Renal disease, no chronic renal failure | 7 | 2 | 3.18 | 1.54–6.55 | 3 | 1 | 4.41 | 0.55–35.14 | |||
Chronic renal failure | 6 | 2 | 4.28 | 1.86–9.83 | 1 | 2 | 0.92 | 0.18–4.71 | |||
No heart disease | 55 | 71 | Ref | 69 | 75 | Ref | |||||
Heart disease, no CHF | 21 | 23 | 1.28 | 0.88–1.86 | 25 | 14 | 1.89 | 1.07–3.35 | |||
CHF | 24 | 6 | 6.36 | 3.74–10.83 | 5 | 10 | 0.58 | 0.27–1.26 | |||
No diabetes | 80 | 88 | Ref | 88 | 87 | Ref | |||||
Diabetes, no complication | 16 | 11 | 1.63 | 1.05–2.53 | 11 | 8 | 1.13 | 0.54–2.34 | |||
Diabetes complications | 4 | 2 | 2.56 | 1.06–6.19 | 1 | 4 | 0.21 | 0.06–0.78 | |||
No lung disease | 51 | 71 | Ref | 71 | 73 | Ref | |||||
Lung disease, no chronic lung disease | 9 | 15 | 0.86 | 0.51–1.43 | 15 | 14 | 1.09 | 0.60–2.00 | |||
Chronic lung disease | 40 | 14 | 4.23 | 2.88–6.22 | 14 | 14 | 1.06 | 0.58–1.95 | |||
Group 3: Functional status variables defined by chart review | Diagnosis of dementia | 21 | 9 | 2.81 | 1.79–4.40 | 8 | 15 | 0.51 | 0.27–0.96 | ||
Requires assistance for bathing | 41 | 8 | 13.43 | 7.63–22.65 | 6 | 14 | 0.48 | 0.24–0.98 | |||
Requires assistance for ambulation | 45 | 15 | 5.45 | 3.69–8.07 | 13 | 22 | 0.63 | 0.36–1.09 | |||
Lives in a non-home setting | 19 | 9 | 2.80 | 1.75–4.49 | 8 | 13 | 0.55 | 0.30–1.01 |
CI, confidence interval; CHF, congestive heart failure.
Severity covariates defined by diagnosis codes (Group 2 variables). To distinguish persons who were assigned diagnosis codes suggestive of more severe illness from those assigned other illness codes, we created additional categories from the groupings of codes used to define the major Group 1 variables. For example, persons meeting the Group 1 cancer definition were subclassified as those with a code for ‘serious’ (leukaemia; lymphoma; or cancer of the brain, oesophagus, liver, lung, pancreas, peritoneum, pleura, or stomach) or metastatic cancer and those with other cancer codes. Similarly, persons meeting the definitions of heart disease, diabetes, renal disease, and lung disease were further categorized as those meeting the more restricted definitions of congestive heart failure, diabetes complications, chronic renal failure, and pneumonia and chronic lung disease, respectively, and the remaining were categorized as those not meeting these more restricted definitions (Group 1 and 2 covariate definitions are available as supplementary data at IJE Online).
Functional status indicators defined by medical record review (Group 3 variables). Information documented in the paper medical record during the 5 years prior to the index date was reviewed to identify whether there was a diagnosis of dementia; whether the subject lived in a non-home setting (such as an assisted living facility); whether they could ambulate without assistance (i.e. did not require a cane, walker, or wheelchair); and whether they required assistance for bathing. To establish baseline status, the functional status variables other than dementia were defined by indications of the person's status prior to the January 1, 1998 start of the outcome period, and dementia was defined by information recorded prior to the index date.
Other variables. Administration of the 1997/1998 influenza vaccine was identified from the Group Health immunization database and by medical record review. Of cases and of controls 3% had an influenza vaccination recorded in the medical record but not in the immunization database. Receipt of home health services during 1997 was identified from claims data.
We hypothesized that subjects with screening laboratory tests performed during outpatient care, and those who sought outpatient optometry services, may represent more homogeneous groups with respect to health status and utilization of health care services than all study subjects. Therefore, we identified study participants with serum creatinine results reported by the Group Health laboratory and those with optometry department visits during 1997.
Analytical methods
Logistic regression was used to assess predictors of vaccination status among controls adjusted for age (5 year age-groups from 65 through 89 and ≥90 years) and sex. Matched unadjusted and covariate-adjusted analyses of the association of influenza vaccination and risk of death were conducted using conditional logistic regression.
Results
Of the 272 cases and 659 matched controls initially identified, chart reviews were completed for 267 (98%) cases and 648 (98%) controls. Of those, ability to ambulate or to bathe without assistance could not be determined for 15 (6%) cases and 22 (3%) controls and those subjects were excluded from the analytical dataset, as were 50 controls that were no longer matched with a case. The final analytical dataset therefore contained 252 cases (persons who died during January through March 1998) and 576 matched controls.
Of those, women represented 48% of cases and 49% of controls, and the proportions for ages 65–74, 75–84, and ≥85 years were 21, 50, and 29% for cases and 21, 50, and 28% for controls, respectively. A record of a 1997/1998 influenza vaccination prior to the index date was identified for 70% of cases and 80% of controls.
The unadjusted odds ratio (OR) for the association of influenza vaccination and risk of death during January through March 1998 was 0.59 (95% CI, 0.41–0.83).
Evaluation of covariates defined by diagnosis codes
As shown in Table 1, the Group 1 diagnosis code covariates, defined by methods used in previously published studies, were in general more common in cases than controls, and among controls these covariates were generally more common in vaccinated persons compared with unvaccinated persons. Adjusting for the Group 1 covariates therefore resulted in a measure of the association of vaccination and risk of death that was further from the null (towards a greater vaccine effect) (OR, 0.45; 95% CI, 0.30–0.68) than the estimate derived from the unadjusted analysis.
The Group 2 covariates shown in Table 1 were created by further stratification of the diagnosis codes used to define the Group 1 variables of cancer, diabetes, heart disease, lung disease, and renal disease. As can be seen by comparing the analysis of the Group 1 binary cancer variable with the Group 2 multiple-category cancer variable, the relationship with risk of death, and with likelihood of vaccination among controls, was markedly different. The presence of any cancer code was associated with an ∼3-fold higher risk of death and a 2-fold higher likelihood of vaccination in comparison with the referent group. In contrast, the presence of a serious or metastatic cancer code was associated with an 11-fold increased risk of death but no difference in likelihood of vaccination, while the presence of a non-serious cancer code was associated with a 3-fold higher likelihood of vaccination but no difference in the risk of death. Similarly, for the other variables, the disease category associated with the highest risk of death (e.g. heart failure or diabetes complications) was associated with, if anything, a lower likelihood of vaccination.
In the analytical model substituting the Group 2 multiple-category covariates for cancer, diabetes, and heart, lung, and renal disease for the corresponding Group 1 binary variables (and retaining the other Group 1 variables), the estimate of the association of influenza vaccination and risk of death was closer to the null (OR, 0.51; 95% CI, 0.33–0.78) than the estimate derived from the Group 1 model (OR, 0.45), though still lower than the crude OR of 0.59.
Comparison of the Group 1 and Group 2 model estimates illustrates the bias in the estimate of the association of vaccination and risk of death produced by adjustment for covariates defined by the broad grouping of diagnosis codes. Those groupings tend to jointly classify as diseased both persons with codes associated with an increased likelihood of vaccination (but not an increased risk of death) and persons with codes associated with an increased risk of death (but not an increased likelihood of vaccination) to create a combined group defined as having both an increased risk of death and, if anything, an increased likelihood of vaccination.
Evaluation of functional status indicators defined by chart review
As shown in Table 1, markers of disability were more prevalent in cases compared with controls. For example, 41% of cases, compared with only 8% of controls, required assistance with bathing. These disability indicators tended to be associated with both a higher risk of death and a decreased likelihood of vaccination. Consequently, adjustment for the functional status indicators moved the estimate of the association of influenza vaccination and risk of death closer to the null (OR, 0.71; 95% CI, 0.47–1.06) than the unadjusted estimate.
Misclassification of health status by diagnosis code variables
As previously demonstrated, adjustment for the Group 2 multiple-category diagnosis code covariates does not appear to properly control for the healthy vaccinee effect. This is likely due to two factors. First, there still may be considerable heterogeneity in illness severity among persons defined by the more restricted Group 2 categories. For example, among persons assigned a diagnosis code for chronic renal disease, one-third had a maximum serum creatinine value in 1997 of <2.4 mg/dl and two-third had higher values. Of cases assigned a chronic renal disease code 85% (11/13) had a creatinine value ≥2.4 mg/dl, compared with 37% (3/8) of controls assigned a chronic renal disease code. The presence of a creatinine value ≥2.4 mg/dl was thus strongly associated with death (OR, 18.11; 95% CI, 6.15–53.27), and was also associated with a trend towards lower likelihood of vaccination among controls (OR, 0.28; 95% CI, 0.04–1.83), but this marker of renal disease severity cannot be defined by diagnosis codes.
Second, a number of persons who are defined as free of comorbidity on the basis of diagnosis code criteria are nonetheless likely to have health impairments predictive of an increased risk of death. This can be seen by evaluating the characteristics of cases and controls who lack any of the diagnosis codes for cancer, dementia, diabetes, or renal, lung, or heart disease, and who would thus be included in the referent group of ‘not diseased’ for all of those covariates (Table 2). Among the cases thus defined as free of comorbidity by diagnosis code criteria, 73% required assistance for ambulation or bathing, lived in a non-home setting, had a diagnosis of dementia identified by chart review, or had received home health services. In contrast, these markers of disability or poor health were identified in only 21% of controls defined as free of comorbidity.
Characteristic . | % ‘not diseased’ cases (n = 34) . | % ‘not diseased’ controls (n = 203) . |
---|---|---|
Diagnosis of dementia identified by chart review | 32 | 3 |
Lives in a non-home setting | 44 | 8 |
Requires assistance for ambulation | 56 | 12 |
Requires assistance for bathing | 32 | 3 |
Received home health services in 1997 | 32 | 8 |
Any of the above | 73 | 21 |
Influenza vaccination | 29 | 78 |
Characteristic . | % ‘not diseased’ cases (n = 34) . | % ‘not diseased’ controls (n = 203) . |
---|---|---|
Diagnosis of dementia identified by chart review | 32 | 3 |
Lives in a non-home setting | 44 | 8 |
Requires assistance for ambulation | 56 | 12 |
Requires assistance for bathing | 32 | 3 |
Received home health services in 1997 | 32 | 8 |
Any of the above | 73 | 21 |
Influenza vaccination | 29 | 78 |
Characteristic . | % ‘not diseased’ cases (n = 34) . | % ‘not diseased’ controls (n = 203) . |
---|---|---|
Diagnosis of dementia identified by chart review | 32 | 3 |
Lives in a non-home setting | 44 | 8 |
Requires assistance for ambulation | 56 | 12 |
Requires assistance for bathing | 32 | 3 |
Received home health services in 1997 | 32 | 8 |
Any of the above | 73 | 21 |
Influenza vaccination | 29 | 78 |
Characteristic . | % ‘not diseased’ cases (n = 34) . | % ‘not diseased’ controls (n = 203) . |
---|---|---|
Diagnosis of dementia identified by chart review | 32 | 3 |
Lives in a non-home setting | 44 | 8 |
Requires assistance for ambulation | 56 | 12 |
Requires assistance for bathing | 32 | 3 |
Received home health services in 1997 | 32 | 8 |
Any of the above | 73 | 21 |
Influenza vaccination | 29 | 78 |
Subgroup analyses
Given the strength of the selection bias in the association of influenza vaccine and risk of all cause mortality, complete adjustment for confounding factors may be difficult. One method to decrease the influence of unmeasured or poorly measured confounders is to restrict the study population to more homogenous subgroups. In all such subgroup analyses reported in Table 3, the unadjusted and adjusted estimates of the association of influenza vaccination and death were closer to the null, and so less biased, than in the analyses of all subjects. In the subgroup of persons with a serum creatinine value recorded, which included 78% of cases, the point estimate of the odds ratio in the model adjusted for the functional status variables (OR, 0.93) was very close to the null, although confidence intervals were wide. A similar effect was found by restricting the analysis to persons who had an optometry visit in 1997, a group that included 39% of cases.
. | . | . | Model . | . | . | . | . | . | |||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
. | . | . | Unadjusted . | . | Adjusted for variables defined by diagnosis codes using previously published methodsa . | . | Adjusted for functional status variablesb . | . | |||||
Population . | No. cases . | No. matched controls . | OR . | 95% CI . | OR . | 95% CI . | OR . | 95% CI . | |||||
All subjects | 252 | 576 | 0.59 | 0.41–0.83 | 0.45 | 0.30–0.68 | 0.71 | 0.47–1.06 | |||||
Subjects who lived in home settings | 203 | 434 | 0.65 | 0.43–0.98 | 0.59 | 0.37–0.96 | 0.74 | 0.46–1.18 | |||||
Subjects who met the criteria for at least one of the Group 1 variables of cancer, diabetes mellitus, dementia, heart disease, lung disease, or renal disease | 218 | 327 | 0.69 | 0.44–1.10 | 0.62 | 0.37–1.03 | 0.82 | 0.48–1.41 | |||||
Subjects with a serum creatinine value recorded in 1997 | 198 | 285 | 0.84 | 0.52–1.38 | 0.84 | 0.47–1.51 | 0.93 | 0.53–1.64 | |||||
Subjects with an optometry department visit in 1997 | 98 | 112 | 0.72 | 0.31–1.63 | 0.35 | 0.12–1.08 | 0.86 | 0.35–2.12 |
. | . | . | Model . | . | . | . | . | . | |||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
. | . | . | Unadjusted . | . | Adjusted for variables defined by diagnosis codes using previously published methodsa . | . | Adjusted for functional status variablesb . | . | |||||
Population . | No. cases . | No. matched controls . | OR . | 95% CI . | OR . | 95% CI . | OR . | 95% CI . | |||||
All subjects | 252 | 576 | 0.59 | 0.41–0.83 | 0.45 | 0.30–0.68 | 0.71 | 0.47–1.06 | |||||
Subjects who lived in home settings | 203 | 434 | 0.65 | 0.43–0.98 | 0.59 | 0.37–0.96 | 0.74 | 0.46–1.18 | |||||
Subjects who met the criteria for at least one of the Group 1 variables of cancer, diabetes mellitus, dementia, heart disease, lung disease, or renal disease | 218 | 327 | 0.69 | 0.44–1.10 | 0.62 | 0.37–1.03 | 0.82 | 0.48–1.41 | |||||
Subjects with a serum creatinine value recorded in 1997 | 198 | 285 | 0.84 | 0.52–1.38 | 0.84 | 0.47–1.51 | 0.93 | 0.53–1.64 | |||||
Subjects with an optometry department visit in 1997 | 98 | 112 | 0.72 | 0.31–1.63 | 0.35 | 0.12–1.08 | 0.86 | 0.35–2.12 |
CI, confidence interval.
Model includes the covariates of atrial fibrillation, cancer, diabetes mellitus, dementia, heart disease, hypertension, lipid disorder, lung disease, renal disease, and rheumatologic disease.
Model includes the functional status covariates of lives in a non-home residence, chart-confirmed diagnosis of dementia, requires assistance for bathing, requires assistance for ambulation.
. | . | . | Model . | . | . | . | . | . | |||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
. | . | . | Unadjusted . | . | Adjusted for variables defined by diagnosis codes using previously published methodsa . | . | Adjusted for functional status variablesb . | . | |||||
Population . | No. cases . | No. matched controls . | OR . | 95% CI . | OR . | 95% CI . | OR . | 95% CI . | |||||
All subjects | 252 | 576 | 0.59 | 0.41–0.83 | 0.45 | 0.30–0.68 | 0.71 | 0.47–1.06 | |||||
Subjects who lived in home settings | 203 | 434 | 0.65 | 0.43–0.98 | 0.59 | 0.37–0.96 | 0.74 | 0.46–1.18 | |||||
Subjects who met the criteria for at least one of the Group 1 variables of cancer, diabetes mellitus, dementia, heart disease, lung disease, or renal disease | 218 | 327 | 0.69 | 0.44–1.10 | 0.62 | 0.37–1.03 | 0.82 | 0.48–1.41 | |||||
Subjects with a serum creatinine value recorded in 1997 | 198 | 285 | 0.84 | 0.52–1.38 | 0.84 | 0.47–1.51 | 0.93 | 0.53–1.64 | |||||
Subjects with an optometry department visit in 1997 | 98 | 112 | 0.72 | 0.31–1.63 | 0.35 | 0.12–1.08 | 0.86 | 0.35–2.12 |
. | . | . | Model . | . | . | . | . | . | |||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
. | . | . | Unadjusted . | . | Adjusted for variables defined by diagnosis codes using previously published methodsa . | . | Adjusted for functional status variablesb . | . | |||||
Population . | No. cases . | No. matched controls . | OR . | 95% CI . | OR . | 95% CI . | OR . | 95% CI . | |||||
All subjects | 252 | 576 | 0.59 | 0.41–0.83 | 0.45 | 0.30–0.68 | 0.71 | 0.47–1.06 | |||||
Subjects who lived in home settings | 203 | 434 | 0.65 | 0.43–0.98 | 0.59 | 0.37–0.96 | 0.74 | 0.46–1.18 | |||||
Subjects who met the criteria for at least one of the Group 1 variables of cancer, diabetes mellitus, dementia, heart disease, lung disease, or renal disease | 218 | 327 | 0.69 | 0.44–1.10 | 0.62 | 0.37–1.03 | 0.82 | 0.48–1.41 | |||||
Subjects with a serum creatinine value recorded in 1997 | 198 | 285 | 0.84 | 0.52–1.38 | 0.84 | 0.47–1.51 | 0.93 | 0.53–1.64 | |||||
Subjects with an optometry department visit in 1997 | 98 | 112 | 0.72 | 0.31–1.63 | 0.35 | 0.12–1.08 | 0.86 | 0.35–2.12 |
CI, confidence interval.
Model includes the covariates of atrial fibrillation, cancer, diabetes mellitus, dementia, heart disease, hypertension, lipid disorder, lung disease, renal disease, and rheumatologic disease.
Model includes the functional status covariates of lives in a non-home residence, chart-confirmed diagnosis of dementia, requires assistance for bathing, requires assistance for ambulation.
Discussion
In this nested case–control study of Group Health seniors, we found that influenza vaccination was associated with a 41% reduction in the risk of all cause mortality during an influenza season. This is consistent with our findings in a large retrospective cohort study of Group Health seniors followed over an 8 year period, which is the cohort from which this nested case–control sample was derived. In that cohort study, we also found a 61% reduction in the risk of all cause mortality in the pre-influenza period, which indicates the presence of a healthy vaccinee bias.
In this case–control study, functional limitations identified by chart review were associated with both an increased risk of death and a decreased likelihood of influenza vaccination. Consequently, adjustment for functional limitations moved the association of influenza vaccination and risk of death towards the null (OR, 0.71), compared with the results of the unadjusted analysis (OR, 0.59). Our results therefore suggest that the functional status limitations identified by chart review are important confounders of the association of influenza vaccination and risk of death in seniors.
In contrast, adjustment for covariates defined by broad groupings of diagnosis codes moved the association away from the null (OR, 0.45), an effect that was less pronounced in the model that included covariates defined by more specific categorizations of diagnosis codes (OR, 0.51). These effects are not consistent with proper adjustment for a healthy vaccinee bias. We sought to better understand why inclusion of the diagnosis code covariates in multivariate models did not properly adjust for bias in the association of influenza vaccination and risk of death in our population. Our results suggest that the diagnosis code covariates do not accurately distinguish the influential subgroup of persons who are both at increased risk of short-term mortality and who are less likely to receive influenza vaccine from among the population of all seniors.
It should be noted that, while diagnosis codes are often used as surrogate markers of disease status, this use is subject to a number of limitations. Diagnosis codes are assigned at medical encounters, which, for our study population, consisted primarily of outpatient visits, and so assignment may vary by medical utilization. In addition, the likelihood that a code for an underlying condition will be assigned at a visit may depend on the reason for the visit, the number of other comorbidities, and physician coding practices. For these reasons, the absence of a diagnosis code is not necessarily a sensitive indicator for the absence of disease,24,25 and our results suggest that, in our population, this misclassification was differential between cases and controls.
Another problem is that there is heterogeneity in disease status among persons classified by broad groupings of diagnosis codes, and even among those classified by more restricted code definitions. We found that health status covariates defined by broad groupings of diagnosis codes classify as diseased both persons with codes indicative of less severe illness, who are not an increased risk of death but who may have a higher likelihood of vaccination, and persons with codes indicative of more severe illness, who may be at increased risk of death but who tend to have a lower likelihood of vaccination. Therefore, the broad groupings of codes do not accurately classify persons according to the confounding factors associated with both risk of death and likelihood of vaccination.
As we showed, even among those assigned the relatively specific diagnosis codes for chronic renal disease there is heterogeneity in the degree of morbidity related to that condition, as evidenced by the creatinine level, which cannot be further differentiated by diagnosis codes. This implies that, if it is important to distinguish persons with severe disease from those with stable disease, as it appears to be when evaluating risk of short-term all cause mortality, covariates defined by diagnosis codes, at least using the methods that we explored, do not properly adjust for confounding by health status.
A limitation of our study is that we evaluated one HMO population in the United States and defined our covariates based on information from those administrative data sources. It is possible that confounding factors of the association of vaccination and risk of death, or indicators of those confounding factors, could vary in other populations. The direction of the effect of adjustment for diagnosis code variables in our analyses is, however, consistent with that reported by other observational studies that adjusted for similar covariates (Table 4).
. | . | Relative risk (95% CI) of the association of influenza vaccine and risk of all cause mortality during influenza season . | . | |
---|---|---|---|---|
Publication author, (year)ref . | Definition of the influenza season outcome period . | Unadjusted for markers of comorbidity . | Adjusted for markers of comorbidity . | |
Nichol (1994)1 | January 1, 1991 to March 31, 1991 | 0.57 | 0.49 | |
(CI not reported) | (0.35–0.70) | |||
November 15, 1991 to March 31, 1992 | 0.56 | 0.46 | ||
(CI not reported) | (0.35–0.61) | |||
December 15, 1992 to March 31, 1993 | 0.64 | 0.61 | ||
(CI not reported) | (0.47–0.81) | |||
Fleming (1995)9 | November 1, 1989 to January 15, 1990 | 0.47 | 0.25 | |
(0.15–1.48) | (0.08–0.79) | |||
Nordin (2001)3 a | October 5, 1996 to May 3, 1997 | 0.58 | 0.51 | |
(CI not reported) | (0.42–0.60) | |||
November 23, 1997 to April 4, 1998 | 0.73 | 0.63 | ||
(CI not reported) | (0.51–0.77) | |||
Nichol (2003)5 | 1998/1999 season—dates not reported | 0.58 | 0.52 | |
(0.53–0.63) | (0.47–0.57) | |||
1999/2000 season—dates not reported | 0.56 | 0.50 | ||
(0.51–0.62) | (0.46–0.55) | |||
Voordouw (2003)10 | December 1, 1996 to June 1, 1997 | 0.88 | 0.76 | |
(0.70–1.10) | (0.60–0.97) | |||
Voordouw (2004)12 | Each calendar year from 1997 through 2002 | 0.91 | 0.78 | |
(0.84–0.99) | (0.72–0.85) | |||
Hak (2005)2 | December 5,1999 to March 4, 2000 | 0.57 | 0.50 | |
(0.43–0.77) | 3(0.32–0.77) |
. | . | Relative risk (95% CI) of the association of influenza vaccine and risk of all cause mortality during influenza season . | . | |
---|---|---|---|---|
Publication author, (year)ref . | Definition of the influenza season outcome period . | Unadjusted for markers of comorbidity . | Adjusted for markers of comorbidity . | |
Nichol (1994)1 | January 1, 1991 to March 31, 1991 | 0.57 | 0.49 | |
(CI not reported) | (0.35–0.70) | |||
November 15, 1991 to March 31, 1992 | 0.56 | 0.46 | ||
(CI not reported) | (0.35–0.61) | |||
December 15, 1992 to March 31, 1993 | 0.64 | 0.61 | ||
(CI not reported) | (0.47–0.81) | |||
Fleming (1995)9 | November 1, 1989 to January 15, 1990 | 0.47 | 0.25 | |
(0.15–1.48) | (0.08–0.79) | |||
Nordin (2001)3 a | October 5, 1996 to May 3, 1997 | 0.58 | 0.51 | |
(CI not reported) | (0.42–0.60) | |||
November 23, 1997 to April 4, 1998 | 0.73 | 0.63 | ||
(CI not reported) | (0.51–0.77) | |||
Nichol (2003)5 | 1998/1999 season—dates not reported | 0.58 | 0.52 | |
(0.53–0.63) | (0.47–0.57) | |||
1999/2000 season—dates not reported | 0.56 | 0.50 | ||
(0.51–0.62) | (0.46–0.55) | |||
Voordouw (2003)10 | December 1, 1996 to June 1, 1997 | 0.88 | 0.76 | |
(0.70–1.10) | (0.60–0.97) | |||
Voordouw (2004)12 | Each calendar year from 1997 through 2002 | 0.91 | 0.78 | |
(0.84–0.99) | (0.72–0.85) | |||
Hak (2005)2 | December 5,1999 to March 4, 2000 | 0.57 | 0.50 | |
(0.43–0.77) | 3(0.32–0.77) |
Estimates of vaccine effectiveness (VE) can be calculated using the formula, VE = 1 − relative risk.
Results of both unadjusted and adjusted analyses were reported by site for the three sites included in the study. For simplicity, only the estimates reported for the largest of the three sites for each of the two study years are included in this table. Similar effects of adjustment were reported in analyses of the other two sites.
. | . | Relative risk (95% CI) of the association of influenza vaccine and risk of all cause mortality during influenza season . | . | |
---|---|---|---|---|
Publication author, (year)ref . | Definition of the influenza season outcome period . | Unadjusted for markers of comorbidity . | Adjusted for markers of comorbidity . | |
Nichol (1994)1 | January 1, 1991 to March 31, 1991 | 0.57 | 0.49 | |
(CI not reported) | (0.35–0.70) | |||
November 15, 1991 to March 31, 1992 | 0.56 | 0.46 | ||
(CI not reported) | (0.35–0.61) | |||
December 15, 1992 to March 31, 1993 | 0.64 | 0.61 | ||
(CI not reported) | (0.47–0.81) | |||
Fleming (1995)9 | November 1, 1989 to January 15, 1990 | 0.47 | 0.25 | |
(0.15–1.48) | (0.08–0.79) | |||
Nordin (2001)3 a | October 5, 1996 to May 3, 1997 | 0.58 | 0.51 | |
(CI not reported) | (0.42–0.60) | |||
November 23, 1997 to April 4, 1998 | 0.73 | 0.63 | ||
(CI not reported) | (0.51–0.77) | |||
Nichol (2003)5 | 1998/1999 season—dates not reported | 0.58 | 0.52 | |
(0.53–0.63) | (0.47–0.57) | |||
1999/2000 season—dates not reported | 0.56 | 0.50 | ||
(0.51–0.62) | (0.46–0.55) | |||
Voordouw (2003)10 | December 1, 1996 to June 1, 1997 | 0.88 | 0.76 | |
(0.70–1.10) | (0.60–0.97) | |||
Voordouw (2004)12 | Each calendar year from 1997 through 2002 | 0.91 | 0.78 | |
(0.84–0.99) | (0.72–0.85) | |||
Hak (2005)2 | December 5,1999 to March 4, 2000 | 0.57 | 0.50 | |
(0.43–0.77) | 3(0.32–0.77) |
. | . | Relative risk (95% CI) of the association of influenza vaccine and risk of all cause mortality during influenza season . | . | |
---|---|---|---|---|
Publication author, (year)ref . | Definition of the influenza season outcome period . | Unadjusted for markers of comorbidity . | Adjusted for markers of comorbidity . | |
Nichol (1994)1 | January 1, 1991 to March 31, 1991 | 0.57 | 0.49 | |
(CI not reported) | (0.35–0.70) | |||
November 15, 1991 to March 31, 1992 | 0.56 | 0.46 | ||
(CI not reported) | (0.35–0.61) | |||
December 15, 1992 to March 31, 1993 | 0.64 | 0.61 | ||
(CI not reported) | (0.47–0.81) | |||
Fleming (1995)9 | November 1, 1989 to January 15, 1990 | 0.47 | 0.25 | |
(0.15–1.48) | (0.08–0.79) | |||
Nordin (2001)3 a | October 5, 1996 to May 3, 1997 | 0.58 | 0.51 | |
(CI not reported) | (0.42–0.60) | |||
November 23, 1997 to April 4, 1998 | 0.73 | 0.63 | ||
(CI not reported) | (0.51–0.77) | |||
Nichol (2003)5 | 1998/1999 season—dates not reported | 0.58 | 0.52 | |
(0.53–0.63) | (0.47–0.57) | |||
1999/2000 season—dates not reported | 0.56 | 0.50 | ||
(0.51–0.62) | (0.46–0.55) | |||
Voordouw (2003)10 | December 1, 1996 to June 1, 1997 | 0.88 | 0.76 | |
(0.70–1.10) | (0.60–0.97) | |||
Voordouw (2004)12 | Each calendar year from 1997 through 2002 | 0.91 | 0.78 | |
(0.84–0.99) | (0.72–0.85) | |||
Hak (2005)2 | December 5,1999 to March 4, 2000 | 0.57 | 0.50 | |
(0.43–0.77) | 3(0.32–0.77) |
Estimates of vaccine effectiveness (VE) can be calculated using the formula, VE = 1 − relative risk.
Results of both unadjusted and adjusted analyses were reported by site for the three sites included in the study. For simplicity, only the estimates reported for the largest of the three sites for each of the two study years are included in this table. Similar effects of adjustment were reported in analyses of the other two sites.
The purpose of our study was to compare the effect of adjustment for markers of functional status with that of adjustment for covariates defined by diagnosis codes, and we did not attempt to obtain a comprehensive classification of illness status by chart review. While the functional status indicators we measured appear to be important factors in the association of influenza vaccination and risk of death, adjustment for those factors probably does not completely control for confounding in the analyses of all subjects. We explored this possibility by restricting the study population to subgroups that may be more homogeneous with respect to health status and health care seeking behaviour. Particularly in the subgroups defined by serum creatinine testing and optometry visits, the analyses adjusting for functional status markers produced estimates of the association of vaccination and risk of all cause mortality that were closer to the null than the estimates from similar analyses of all subjects. Our findings indicate that further research is needed on methods to reduce the influence of bias on estimates of influenza vaccine effectiveness in observational studies.
Functional status limitations may confound the association of influenza vaccination and risk of all cause mortality in seniors, but these factors are not captured in the administrative data sources used in many of the published evaluations of influenza vaccine effectiveness.
We conducted a nested case–control study of persons ≥65 years of age which included 252 cases who died during an influenza season and 576 age-matched controls and identified functional limitations, such as requiring assistance for bathing, by chart review.
Functional limitations were important confounders of the association of influenza vaccination and risk of all cause mortality, and adjustment for functional limitations moved the association of influenza vaccination and risk of death towards the null (OR, 0.71), compared with the results of the unadjusted analysis (OR, 0.59).
These findings suggest that assessment of functional status limitations should be considered in the design of observational studies of influenza vaccine effectiveness in seniors.
References
Nichol KL, Margolis KL, Wuorenma J, Von Sternberg T. The efficacy and cost effectiveness of vaccination against influenza among elderly persons living in the community.
Hak E, Buskens E, van Essen GA et al. Clinical effectiveness of influenza vaccination in persons younger than 65 years with high-risk medical conditions: the PRISMA study.
Nordin J, Mullooly J, Poblete S et al. Influenza vaccine effectiveness in preventing hospitalizations and deaths in persons 65 years or older in Minnesota, New York, and Oregon: data from 3 health plans.
Nichol KL, Baken L, Nelson A. Relation between influenza vaccination and outpatient visits, hospitalization, and mortality in elderly persons with chronic lung disease.
Nichol KL, Nordin J, Mullooly J, Lask R, Fillbrandt K, Iwane M. Influenza vaccination and reduction in hospitalizations for cardiac disease and stroke among the elderly.
Shapiro Y, Shemer J, Heymann A et al. Influenza vaccination: reduction in hospitalizations and death rates among members of “Maccabi Healthcare Services” during the 2000–2001 influenza season.
Fedson DS, Wajda A, Nicol JP, Hammond GW, Kaiser DL, Roos LL. Clinical effectiveness of influenza vaccination in Manitoba.
Nichol KL, Wuorenma J, von Sternberg T. Benefits of influenza vaccination for low-, intermediate-, and high-risk senior citizens.
Fleming DM, Watson JM, Nicholas S, Smith GE, Swan AV. Study of the effectiveness of influenza vaccination in the elderly in the epidemic of 1989–90 using a general practice database.
Voordouw BC, van der Linden PD, Simonian S, van der Lei J, Sturkenboom MC, Stricker BH. Influenza vaccination in community-dwelling elderly: impact on mortality and influenza-associated morbidity.
Hak E, Wei F, Grobbee DE, Nichol KL. A nested case–control study of influenza vaccination was a cost-effective alternative to a full cohort analysis.
Voordouw AC, Sturkenboom MC, Dieleman JP et al. Annual revaccination against influenza and mortality risk in community-dwelling elderly persons.
Schade CP, McCombs MA. Influenza immunization and mortality among diabetic Medicare beneficiaries in West Virginia.
Mangtani P, Cumberland P, Hodgson CR, Roberts JA, Cutts FT, Hall AJ. A cohort study of the effectiveness of influenza vaccine in older people, performed using the United kingdom general practice research database.
Landi F, Onder G, Cesari M et al. Effects of influenza vaccination on mortality among frail, community-living elderly patients: an observational study.
Heymann AD, Shapiro Y, Chodick G et al. Reduced hospitalizations and death associated with influenza vaccination among patients with and without diabetes.
Wang ST, Lee LT, Chen LS, Chen TH. Economic evaluation of vaccination against influenza in the elderly: an experience from a population-based influenza vaccination program in Taiwan.
Christenson B, Lundbergh P, Hedlund J, Ortqvist A. Effects of a large-scale intervention with influenza and 23-valent pneumococcal vaccines in adults aged 65 years or older: a prospective study.
Jackson LA, Neuzil KM, Yu O et al. Effectiveness of pneumococcal polysaccharide vaccine in older adults.
Pressley JC, Patrick CH. Frailty bias in comorbidity risk adjustments of community-dwelling elderly populations.
Covinsky KE, Justice AC, Rosenthal GE, Palmer RM, Landefeld CS. Measuring prognosis and case mix in hospitalized elders. The importance of functional status.
Inouye SK, Bogardus ST Jr, Vitagliano G et al. Burden of illness score for elderly persons: risk adjustment incorporating the cumulative impact of diseases, physiologic abnormalities, and functional impairments.
Inouye SK, Peduzzi PN, Robison JT, Hughes JS, Horwitz RI, Concato J. Importance of functional measures in predicting mortality among older hospitalized patients.
Pressley JC, Trott C, Tang M, Durkin M, Stern Y. Dementia in community-dwelling elderly patients: A comparison of survey data, medicare claims, cognitive screening, reported symptoms, and activity limitations.