Derivation of populations for group 1, data sets for group 1, and group 2. MDS indicates Minimum Data Set. *All residents who had any correction submitted were eliminated, resulting in 0.31% of assessments removed. †All residents who had any correction submitted were eliminated, resulting in 0.30% of assessments removed. ‡Because many residents had more than 1 criterion that would make them ineligible, the exclusions were applied sequentially (thus, these numbers reflect residents eliminated at each step so that no resident is counted in >1 exclusion category). §Variables included activities of daily living, cognition using components of the Cognitive Performance Scale,32 constipation, delusions, depression using components of the depression scale developed by Burrows et al,33 fever, sex, hallucinations, urinary tract infection, respiratory tract infection, pain using the scale developed by Fries et al,34 and physical aggression.
Leonard R, Tinetti ME, Allore HG, Drickamer MA. Potentially Modifiable Resident Characteristics That Are Associated With Physical or Verbal Aggression Among Nursing Home Residents With Dementia. Arch Intern Med. 2006;166(12):1295-1300. doi:10.1001/archinte.166.12.1295
Physical aggression by nursing home residents is a burden to residents and staff. The identification of modifiable correlates would facilitate developing preventive strategies. The objectives of the study were to determine potentially modifiable resident characteristics that are associated with physical aggression and to correlate these characteristics with verbal aggression.
This was a cross-sectional study of nursing home residents in 5 states who had at least 1 annual Minimum Data Set assessment completed during 2002. Case subjects were defined as nursing home residents 60 years and older with dementia who were reported to have been physically aggressive in the week before their assessment. Control subjects were all other residents 60 years and older with dementia. The main outcome measure was being physically aggressive during the past week.
A total of 103 344 residents met study criteria, of whom 7120 (6.9%) had been physically aggressive in the week before their annual Minimum Data Set assessment. After adjustment for potential confounders, including age, sex, severity of cognitive impairment, and dependence in activities of daily living, physical aggression was associated with depressive symptoms (adjusted odds ratio [AOR], 3.3; 99% confidence interval [CI], 3.0-3.6), delusions (AOR, 2.0; 99% CI, 1.7-2.4), hallucinations (AOR, 1.4; 99% CI, 1.1-1.8), and constipation (AOR, 1.3; 99% CI, 1.2-1.5). Urinary tract infections, respiratory tract infections, fevers, reported pain, and participation in recreational activities were not significantly associated with physical aggression in multivariate analyses (P>.01 for all). Except for constipation, the correlates of verbal aggression were similar to those of physical aggression.
If the associations we have estimated are causal, then treatment of depression, delusions, hallucinations, and constipation may reduce physical aggression among nursing home residents.
Aggression by nursing home residents can cause physical and psychological trauma to residents and staff. Nationally, 88 000 (6.8%) of the approximately 1.3 million residents in US nursing homes are physically aggressive each week.1
Dementia2- 15 and psychosis5,16- 19 have been linked to increased physical aggression in almost all studies. There are conflicting findings about the association of physically aggressive behavior with sex,3,5- 9,11,13,16,20- 22 depression,6- 8,13,15,16 level of activities of daily living (ADL) dependence,2,3,5,7,9- 11,14,16,17,20,23- 27 and reported pain.25,28 Few prior studies have used a standardized definition of physical aggression and had enough residents to control for multiple resident characteristics simultaneously or to generalize their findings beyond one nursing home.
Although many studies have examined verbal aggression, they have often subsumed this as agitation or aggression along with heterogeneous behaviors (including physical aggression), making it difficult to distinguish the cause of each behavior. Among those that have explicitly stated verbal aggression as an entity, most studies have indicated that cognitive impairment is a risk factor,6,7,29,30 but there are conflicting results about depression.15,29
The aim of this cross-sectional study was to determine whether a set of potentially modifiable resident characteristics was associated with physical aggression. A secondary aim was to determine whether this same set of characteristics was associated with verbal aggression.
The Minimum Data Set (MDS) is a health assessment that must be completed for all residents of nursing homes that receive federal funds. The MDS includes questions about a resident's medical conditions and functional status. There is a short (3-page) and comprehensive (8-page) version; the former is completed every 3 months unless a comprehensive one has been completed during that period. The comprehensive assessment must be done when there is a significant change in the resident's status (decline or improvement) or at least yearly. A Centers for Medicare & Medicaid Services–sponsored multistate study31 showed that many items on the MDS have high interobserver reliability among nurse researchers.
Potential participants included all 1 095 306 persons residing in a nursing home during 2002 in 1 of 5 large geographically diverse states: California, New York, Ohio, Pennsylvania, or Texas. Inclusion criteria were having had at least 1 comprehensive MDS assessment (annual or done because of a significant change in status), being 60 years or older, and having dementia. Exclusion criteria were having had any correction to an MDS submitted (to eliminate the possibility of uncorrected variables of interest) or being in a coma. Because it was not known whether the prevalence of physical aggression as documented on assessments completed because of a significant change in status would differ from the prevalence documented on those done to fulfill the federal requirement that at least one be done annually, we created 2 groups for study: group 1, in which the first comprehensive annual assessment of 2002 was used (n = 205 284); and group 2, in which the first annual assessment of 2002 or the first comprehensive assessment done due to a significant change was used (n = 306 045). Groups 1 and 2 have some overlap in composition that may be summarized as follows: those residents who had an annual evaluation but then experienced a significant change in status will undergo at least 2 evaluations and will be found in groups 1 and 2, but residents who had a significant change in status before they were scheduled for an annual assessment will be counted in group 2 only (because they would not then undergo an annual assessment in 2002). Group 1, constituting the primary analysis, was further divided into a training data set (n = 103 344) and a validation data set (n = 101 940) by randomly assigning half of the facilities to each data set. The residents in the training data set were used to derive a statistical model, which was then tested in the validation data set and in group 2. These same groups were used for the secondary outcome of verbal aggression. Less than 2% of the original sample was eliminated because of missing data. The Figure shows how each study group was derived.
Because the comprehensive assessments contain more variables of interest for this study than the short assessments, we wanted to use them for this study but were concerned that a resident's physical or verbal aggression on a comprehensive assessment might not be representative of the resident's physical or verbal aggression on other assessments. To determine how representative the comprehensive MDS is of resident aggression throughout the year, we calculated κ35 scores between the presence of physical aggression (dichotomized as 0 or ≥1 event in the prior week) in the first annual MDS and a dichotomized average score (ie, 0 or 1, corresponding to less than or greater than half) in all other assessments (short or comprehensive) done on each resident that year for group 1, and found a κ score of 0.62. We repeated this in an analogous manner to calculate a κ score for group 2 and found a κ score of 0.57. For verbal aggression, the κ scores were 0.67 and 0.63, respectively. Given the high κ scores, we believed the comprehensive MDS assessment was sufficiently representative of residents' average level of physical and verbal aggression.
The main outcome measure was physical aggression, as defined by MDS item “others were hit, shoved, scratched, sexually abused.” The response to this item has 4 answers based on the frequency of aggression during the week before the assessment and when dichotomized as no aggression vs 1 or more episodes has a κ score between nurse researchers and nursing staff of 0.631 (also V. Mor, PhD, written communication, 2003). The secondary outcome measure was verbal aggression, “others were threatened, screamed at, cursed at,” which also has 4 responses and a weighted κ score of 1.0 among nurse researchers.31
We a priori sought to determine whether psychosis (delusions or hallucinations), depression, constipation, decreased participation in recreational therapy, respiratory or urinary tract infections, fever, and pain were associated with physical or verbal aggression. These correlates are potentially amenable to behavioral or pharmacological interventions; many have been examined in smaller studies using non-MDS instruments and were of interest based on our clinical experience.
The corresponding MDS items for these potentially modifiable resident characteristics were as follows: hallucinations; delusions; constipation; pain burden using a 4-level composite score34 based on pain frequency and intensity; urinary tract infection; fever; respiratory tract infection, which we designated as having had either a respiratory tract infection or pneumonia; average amount of awake time involved in activities, excluding treatment or ADL care; and depression, defined as a score of 3 or greater33 and is the frequency sum of making negative statements, persistent anger with self or others, expressing unrealistic fears, repetitive health complaints, repetitive anxious complaints, being sad, pained facial expression, crying, and tearfulness. For all of these, the κ score was 0.65 or greater, except for delusions and constipation, which do not have published values.31
Covariate factors were those associated with physical aggression in previous studies that might confound the potentially modifiable factors and are not themselves modifiable. The covariates included age; sex; cognitive impairment using the Cognitive Performance Scale score,32 which is based on questions about short-term memory, cognitive skill for daily decision making, presence of coma (which was not applied in this study because it was part of the exclusion criteria), being understood by others, and eating self-sufficiency; and ADL, including self-sufficiency in eating, personal hygiene, self-mobility on the unit, dressing self-sufficiency, and toileting self-sufficiency. For all of these, the κ score was 0.63 or greater.31
All of the previously described potential modifiable correlates apply to the same 7 days before the assessment date, with the exception of questions about constipation, which apply to the prior 14 days, and those about urinary tract infections and mood, which apply to the prior 30 days. All questions are completed, except those about infection because infections are only documented if the nurse thinks they are relevant to the resident's ADL, cognitive, mood, or behavior status.
This study was approved by the Human Investigations Committee of Yale Medical School and the Centers for Medicare & Medicaid Services.
We built 2 multivariable models using unconditional logistic regression, accounting for clustering by facility with the Huber-White estimate of variance,36 one for the primary outcome of physical aggression and another for the secondary outcome of verbal aggression. We included all of the previously noted potentially modifiable resident characteristics and controlled for all covariates in both models. The only continuous variable was age, for which we created 10-year intervals with the referent as the most populous (subjects aged 80-89 years). We created indicator variables as follows: Cognitive Performance Scale scores32 were defined as mild (score, 2), moderate (score, 3 or 4), or severe (score, 5 or 6) impairment, with very mild (score, 1) as the referent; pain burden34 was entered as a score of 1, 2, or 3, with no pain as the referent; a variable was included for each of the possible responses for time in recreational activities, with the most active level as the referent; ADL variables for eating, dressing, toileting, and hygiene were generated by making 3 levels for each domain and using independence as the referent; we included 4 levels for locomotion because “did not occur” was included; and the other variables were dichotomized as either present or absent (depression based on an aggregate score,33 respiratory tract infection, fever, female sex, urinary tract infection, constipation, delusions, and hallucinations).
For analysis involving residents in group 1, we used only the first annual assessment done in 2002. For analysis of residents in group 2, we used the first assessment done in 2002, regardless of whether it was a scheduled annual assessment or one done because of a significant change in status. Because of concern that a response of did not occur for self-performance of eating, dressing, toileting, and hygiene may have been somewhat inconsistent, we excluded residents with such responses (1394/103 344 [1.35%]) from the final derivation model and found that the difference in the point estimates was less than 5% compared with a derivation model in which those residents were included.
We assessed the fit of the physical aggression model derived from the training data set by calculating the area under the receiver operating characteristic curve based on the coefficients of the training data set applied to the training data set, the validation data set, and group 2.
All statistical tests are 2-sided, with P<.01 indicating statistical significance and clinical significance defined as an adjusted odds ratio greater than 1.25 or less than 0.8. All analyses were performed with Stata, version 8.0 (Stata Corp, College Station, Tex).
The mean age of residents in the training data set was 84 years; 75.9% were women, and 6.9% were physically aggressive at least once in the week before their assessment. Table 1 shows the number, percentages, and bivariate and multivariable odds ratio estimates of each potentially modifiable characteristic in the training data set.
Among the potentially modifiable factors, the presence of depression had the highest odds of physical aggression. The presence of delusions, hallucinations, and constipation was also associated with increased odds of aggressive behavior. Respiratory tract infections (P=.70), urinary tract infections (P=.86), fever (P=.045), reported pain (mild, P=.02; moderate, P=.0; high, P=.43), and the amount of time participating in non-ADL or treatment activities such as recreational activities (moderate, P=.99; low, P=.28; none, P=.06) did not have a statistically significant association with physical aggression.
The findings among the training data set, validation data set, and group 2 were quite similar, with less than 1% difference in their receiver operating characteristic curve areas (0.74, 0.73, and 0.73, respectively) when the coefficients derived from the training data set were applied to the training set, validation data set, and group 2 with regard to physical aggression.
Approximately 10.5% of the residents were verbally aggressive at least once in the week before their assessment and, similar to physical aggression, statistically significant associations included depression, delusions, and hallucinations. Constipation was not associated with verbal aggression. No other characteristics were significant (Table 2).
There was also some coexistence of these behaviors: 58.4% of patients who were physically aggressive at least once in the week before their assessment were also verbally abusive at least once that week, and 38.4% of patients who were verbally abusive at least once in the week before their assessment were also physically abusive at least once that week.
Depression had the strongest association with aggressive behavior, followed by delusions, hallucinations, and constipation. Urinary tract infection, respiratory tract infection, fever, reported pain, and participation in non-ADL activities were not associated with aggressive behavior.
Of 6 studies in which the association was investigated, 3 found that depression had a positive association7,13,15 with physically aggressive behavior.6,8,16 The MDS scale we used for depression diagnosis has a sensitivity of 91% and a specificity of 69% when compared with assessment by a psychiatrist.33 Our findings may differ from those of some studies because other instruments have considered “hurting oneself”6 as physically aggressive, whereas the MDS's definition of physical aggression applies only to physical contact toward others. Among published pharmacologic trials37- 42 that have sought to treat physical aggression in nursing home residents, none have included an antidepressant; thus, the relative importance of depression as a potentially modifiable risk factor may not be fully recognized. In addition, given the recent Food and Drug Administration black box warning about the increased mortality associated with use of atypical antipsychotics in the treatment of behavioral problems in older patients with dementia,43 some prior studies37,39,41 may no longer be relevant, but studies of antidepressants may be useful.
Our findings that delusions and hallucinations are associated with physically aggressive behavior are consistent with 45,17- 19 of 516 other studies that have looked at psychosis in some form.
To our knowledge, no prior studies have investigated the association of constipation. We chose to study constipation a priori because it is common, modifiable, and recognized by clinicians to be a cause of many nonspecific symptoms. It is not clear whether physical aggression may be related to factors that predispose to constipation (eg, anticholinergic medications such as tricyclic antidepressants), the symptoms associated with constipation, or interventions such as suppositories that may elicit a defensive action by some residents.
We did not find that pain had a statistical association with physical aggression, and prior studies25,28 have found conflicting results. The lack of association persisted even when we analyzed frequency and severity separately.
Participation in non-ADL activities, such as recreational therapy, was not statistically associated with physical or verbal aggression. Although the crude odds ratios suggested that minimal activity was associated with increased physical aggression, this was likely confounded by other covariates because the adjusted odds ratio approached 1 in the adjusted model.
Many researchers have attempted to distinguish verbal and physical aggression. We agree that the distinction is important from an intervention perspective, but our findings show that those with physical or verbal aggression may have the same potentially modifiable characteristics.
Our study has several limitations. First, κ scores were not available for constipation or delusions. Second, with the exception of the physical aggression outcome, the presented κ scores were those for nurse researchers, not for nursing home staff or between nursing home staff and nurse researchers, so we do not know how reliably the nurses reported the characteristics in this study. Third, because the MDS does not specify who the target of the aggressive acts was, we cannot clarify what portion were toward staff, other residents, or visitors, or whether the risk characteristics varied by recipient of the aggression. Fourth, this study did not include facility characteristics. Although some of this information is collected in the Online Survey Certification and Reporting System, we chose to focus on potentially modifiable resident characteristics for this initial study, so we cannot comment on facility characteristics or on whether the relationship between the resident characteristics and aggression is independent of facility characteristics, but by randomizing by facility and accounting for clustering by facility in the analyses we attempted to mitigate bias due to cluster-correlated resident data within facilities. Fifth, validation of the questions about delusions, hallucinations, constipation, non-ADL activities, and infections (urinary tract and respiratory tract) is not available. Sixth, residual confounding for uncontrolled variables or bias due to measurement error is possible with any observational study, including this one. Finally, because this was a cross-sectional observational study and the MDS does not specify a sequence of onset for outcome and exposure variables, we cannot assert a temporal link between resident characteristics and either physical or verbal aggression.
Despite these limitations, to our knowledge this study is the first to use the MDS data as dependent variables; therefore, the findings are directly applicable to the US nursing home population and the Resident Assessment Protocols, which are MDS-based guides intended to help nursing staff develop care plans. We are not aware of a prior study that had a single model in which mutable factors (depression, delusions, and hallucinations) were adjusted for known confounders (age, cognitive impairment, and sex). This study had the size to control, simultaneously, for many relevant confounding variables and included thousands of facilities in different states, thereby increasing the generalizability of the results and showing with substantial precision the relative impact of each independent variable. Furthermore, the model coefficients were similar between those in the validation data set of group 1 (in which the first annual MDS assessment was used) and in those of group 2 (in which the first comprehensive assessment [whether done to satisfy an annual requirement or because of a significant change in a resident's status] was used), further suggesting that these associations are robust.
Physical or verbal aggression among nursing home residents with cognitive impairment may be a major cause of distress among staff and other residents injured by the aggressor, as well as to the aggressor. We found that aggressive behavior among residents was associated with depression, delusions, and hallucinations, and that physical aggression was also associated with constipation. All of these factors may be amenable to intervention and, in addition to reducing the morbidity associated with these entities themselves, effective treatment may reduce the risk of violence in nursing homes.
Correspondence: Ralph Leonard, MD, MPH, CALM-MD, LLC, PO Box 26516, St Louis Park, MN 55426.
Accepted for Publication: March 4, 2006.
Author Contributions: Dr Leonard had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Financial Disclosure: None.
Funding/Support: This study was supported in part by grant T32AG19134 from the National Institute on Aging (Dr Leonard), the Donald W. Reynolds Foundation (Dr Drickamer), and grant P30AG021342 from the Claude D. Pepper Older Americans Independence Center at Yale (which is supported by the National Institute on Aging).
Role of the Sponsor: The funding bodies had no role in data extraction and analyses, in the writing of the manuscript, or in the decision to submit the manuscript for publication.
Previous Presentations: This study was presented in part at the Fourth Annual Senior Symposium of the Connecticut Chapter of the American Society of Consultant Pharmacists; April 16, 2004; Mashuntucket, Conn; and to the Massachusetts Extended Care Federation; May 12 and 26, 2004; Springfield and Dedham, Mass.
Acknowledgment: We thank the faculty and fellows who attended the Yale Geriatrics Research in Progress Meeting for their insightful comments; and the peer reviewers for their thoughtful comments.