A Multifactorial Intervention to Reduce the Risk of Falling among Elderly People Living in the Community

Setting and Subjects

The details of the study design and methods have been reported elsewhere and are summarized here¹⁹. The potential subjects were the 2522 enrollees of a health maintenance organization (HMO) in southern Connecticut who were at least 70 years of age. Sixteen of the 17 eligible physicians who cared for at least 100 of these enrollees agreed to participate. For reasons previously described, the 16 physicians were frequency-matched into four groups of 4 physicians each, on the basis of their high or low scores on two measures -- namely, the number of people at least 70 years of age among their patients (>150 vs. ≤ 150) and the mean number of new prescriptions written per office visit (≥ 1 vs. <1)¹⁹. Two physicians in each group of four were assigned randomly to the control group, and two to the intervention group. Enrolled subjects were assigned to the same study group as their physicians.

The eligibility criteria, in addition to age, included independent ambulation, residence outside a nursing home (i.e., in the community), no current enrollment in another study of aging, a score of at least 20 on the Folstein Mini-Mental State Examination,²⁰ no participation in vigorous sports or walking for exercise within the month before enrollment, and at least one of the risk factors described below. Enrollment took place between October 1, 1990, and April 30, 1992. By means of a random-number generator in SAS software on a PC computer, each patient was assigned a priority number within the physician's practice²¹. Patients were screened sequentially until approximately 20 eligible and consenting subjects were enrolled from each physician's practice. Of the 2229 potential subjects selected randomly, 1950 (87 percent) agreed to be screened. A total of 355 screened subjects (18 percent) met the eligibility criteria. Among these eligible subjects, 301 (85 percent) agreed to participate; 153 subjects were assigned to the intervention group, and 148 to the control group. The eligible patients who declined to participate did not differ significantly from the enrolled subjects in terms of age, sex, or group assignment.

Assessment and Intervention

Table 1. Table 1. Targeted Risk Factors and Corresponding Interventions.

The development of the assessment and intervention protocols has been described in detail elsewhere¹⁹. The risk factors were selected on the basis of evidence of their association with the risk of falling and the availability of assessment measures and interventions considered feasible in usual clinical practice. The risk factors we targeted and the corresponding interventions are shown in Table 1.

The base-line assessments were conducted in the subjects' homes by the study nurse practitioner and physical therapist, who were unaware of the group assignments. The nurse practitioner obtained demographic data, a history of falls, and information on depressive symptoms,²² the presence of chronic diseases, and the level of independence in activities of daily living²³ and administered the Falls Efficacy Scale, a measure of the subject's degree of confidence in performing 10 common activities (such as walking and stair climbing) without falling,²⁴ and the ambulation and mobility subscales of the Sickness Impact Profile²⁵. She also assessed corrected near vision²⁶ and hearing²⁷. The names and dosages of all prescription and nonprescription medications were recorded from the containers. The number of hazards for falling was determined by a room-by-room examination of walking paths, furniture, and stairs.

Within one week of the nurse practitioner's assessment, the physical therapist visited the subjects to assess the risk factors listed in Table 1. Strength and joint impairment were identified by manual muscle testing and tests of range of motion, respectively^28,29. The assessments of balance and transfer skills³⁰ involved observing the subjects for instability while they were sitting, moving to and from a chair or bed, standing, carrying objects, bending over, and reaching. Deviation from a path, missed steps, step height and length, stability in turning, trunk position, and appropriate use of walking aids were observed while the subjects walked 6.1 m (20 ft) on flat and uneven surfaces.

These assessments were repeated for 250 of the 301 participants (83 percent) a median of 4.5 months after the base-line assessment. The staff members performing the reassessments did not know the subjects' group assignments.

Intervention Group

The subjects assigned to the intervention group received the interventions listed in Table 1 as indicated in the base-line assessment^31-34. Decision rules and priority lists were used to select standardized intervention protocols for each identified risk factor^19,30. Recommendations to adjust medications were discussed with the subjects' primary physicians, who then made the final decisions.

Home visits for physical therapy involved gait or transfer-skill training, if needed, as well as instruction in the progressive, competency-based balance and strengthening exercise programs³⁰. Simple illustrated instructions with large print were provided for each exercise program. Subjects were instructed to perform the exercises twice a day for 15 to 20 minutes per session.

The intervention phase lasted three months after the base-line assessment but was extended if the subjects had health problems that interfered temporarily with their ability to exercise. The maintenance phase, during which the study staff contacted the subjects monthly, lasted from the end of the intervention phase until six months after enrollment. Adherence to the exercise programs, as reported by the subjects, was assessed by the physical therapist on a weekly basis.

Control Group

The subjects assigned to the control group received home visits from social-work students, during which structured interviews were conducted^19,35. The number of social visits was matched to the estimated number of visits by a nurse practitioner or physical therapist that would be required for subjects in the intervention group who had comparable risk factors.

Outcome

The primary outcome we studied was the incidence of falls,²⁴ as recorded on a calendar that subjects mailed to the research staff monthly. The subjects were contacted by an interviewer who was blinded to their group assignments if no calendar was returned or if a fall was indicated for any day of the month¹¹. During a follow-up telephone interview, the subjects were asked about medical care sought after falls and injuries sustained. Serious injuries included fractures, head injuries requiring hospitalization, joint dislocations or severe sprains, and lacerations requiring suturing²⁴. Monitoring for falls and injuries began on the day of the physical therapist's base-line assessment. Data on falls for one year of follow-up (or up to the occurrence of a fall before loss to follow-up) were available for 147 subjects in the intervention group and 144 in the control group. Reasons for the 10 losses to follow-up included death (n = 8), severe illness (n = 1), and a decision by the subject not to continue in the study (n = 1).

Statistical Analysis

Base-line characteristics were compared in the two study groups. The length of time to the first fall was compared in the two groups with a log-rank test. The proportion of subjects in each group who fell was compared by means of a crude relative risk. Adjusted relative risks were calculated with use of a generalized linear interactive model³⁶ by modeling the logarithmically transformed outcome probabilities as a linear function of the covariates^37,38. Tests of the interaction of covariates with group assignment were also examined with this model. Since it was the physicians, rather than the subjects, who were assigned randomly to the treatment groups, we tested whether our results might be due to the observed pattern of physician assignment, using a randomization test³⁹. Within each block of four physicians, all permutations of physicians' assignment to the intervention and control groups were obtained (n = 1296); the associated relative risks were calculated and ranked from low to high. The calculated relative risk for the physician-assignment pattern observed in our study was contained in the lowest 5 percent of this normal distribution of relative risks, indicating that our findings were not due merely to the way in which physicians were randomly assigned. Therefore, all further analyses were conducted with the study subjects used as the unit of analysis.

The total incidence density -- that is, the total number of falls divided by the total number of person-weeks of follow-up -- was calculated as a measure of the rate of falls within each group and used to obtain crude incidence-rate ratios. Adjusted incidence-rate ratios were calculated by means of a pooled logistic-regression model^40,41 (with SAS software on a PC computer²¹). Each week of follow-up was assumed to represent an independent period of observation and was included separately for each subject. For each week, the outcome was defined as the occurrence of a fall or no fall, and base-line as well as time-dependent characteristics were included as covariates. To account for the interdependence of observations, dummy variables that modeled the number of previous falls (0, 1, 2, or ≥ 3) and the week of the study (1 through 52) were included as potential covariates.

For subjects who had reassessment interviews, we examined the effect of the intervention on each of the eight risk factors by means of a chi-square test; changes in the total number of risk factors, the environmental-hazard score, and the score on the Falls Efficacy Scale were examined by analysis of covariance, with the base-line score and the study group as independent variables. All P values are two-tailed.

Table 2. Table 2. Base-Line Characteristics of the Subjects, According to Treatment Group.

The base-line characteristics of the subjects in the two groups were similar (Table 2), although a slightly higher proportion of the intervention group than of the control group had education beyond high school, and a lower proportion had impairment of leg strength. The mean number of risk factors per subject was similar in the two groups.

Figure 1. Figure 1. Cumulative Percentages of Subjects in the Intervention and Control Groups Who Had One or More Falls during One Year of Follow-up.

The difference between the groups was significant (P = 0.05, by the log-rank test). The numbers still at risk for a fall at 3, 6, 9, and 12 months are shown below the figure. Only 10 of the subjects were lost to follow-up: 6 in the intervention group and 4 in the control group. Data on subjects were censored after a fall. The cumulative relative risks are shown for 3, 6, 9, and 12 months of follow-up.

Table 3. Table 3. Incidence of Falls and Other Events during One Year of Follow-up, According to Treatment Group.

There was a significant difference between the intervention and control groups in the length of time to the first fall (P = 0.05) (Figure 1) and in the proportion of subjects who fell (35 percent vs. 47 percent, P = 0.04) (Table 3); the crude incidence-rate ratio for falls in the intervention group as compared with the control group was 0.64 (95 percent confidence interval, 0.49 to 0.83) (Table 3). The risk reduction associated with the intervention was maintained in multivariate analyses in which we adjusted for age, sex, previous falls, and number of risk factors, as well as the week of follow-up for the incidence-rate ratio (Table 3). Although the numbers were small, a similar trend toward a reduction in risk in the intervention group was seen for falls requiring medical care or resulting in serious injury (Table 3). Seven subjects in the control group (5 percent) and four subjects in the intervention group (3 percent) had fractures.

The mean (±SD) change in the scores on the Falls Efficacy Scale, on which higher scores indicate greater self-confidence, differed significantly (P = 0.02) between the control group (-1.2 ±4.9) and the intervention group (+0.2 ±3.9), with adjustment for the base-line score. There was no significant difference between the two groups in the mean change in the Sickness Impact Profile scores.

Targeted Intervention Process

The subjects in the intervention group received a mean of 7.8 ±4.0 home visits (range, 0 to 22), as compared with 6.2 ±3.8 (range, 0 to 14) for the control subjects (t = 3.62, P<0.001). Eleven subjects in the intervention group did not complete the intervention phase, including eight who declined to do so and three in whom intervening illnesses developed.

Table 4. Table 4. Subjects in the Intervention Group Who Received Interventions and Types of Interventions Received.

Table 4 shows the number of subjects in the intervention group with each risk factor and the types of interventions they received. The most common reason for not receiving an intervention for a risk factor was the presence of factors with a higher priority or a contraindication to the intervention.

Subjects in the intervention group reported performing 73 percent of the recommended exercise sessions, for a mean of 10 sessions each week. Sixty-five percent of these subjects performed at least 70 percent of sessions; 85 percent of subjects performed over half of the exercise sessions.

Table 5. Table 5. Subjects with Each Risk Factor at Base Line Who Had the Risk Factor at Reassessment.

Risk factors were reassessed for 131 members (86 percent) of the intervention group and 119 members (80 percent) of the control group a median of 4.5 months after the base-line assessment. The percentage of subjects in the intervention and control groups who had a risk factor at base line and continued to have it at reassessment is shown in Table 5. At reassessment, a significantly smaller percentage of the intervention group than of the control group continued to use at least four prescription medications, to transfer unsafely to bathtub or toilet, or to have impairment in balance or gait. Overall, the intervention group had a mean decline of 1.1 ±1.6 in the total number of risk factors, as compared with a decrease of 0.6 ±1.4 in the control group (P = 0.03). When the change in the number of risk factors was added to the model predicting the occurrence of falls among the 242 subjects with complete reassessment data, treatment assignment was no longer significant, and the risk of falling declined by 11 percent (adjusted relative risk, 0.89; 95 percent confidence interval, 0.79 to 1.00) for each decrease of 1.0 in the number of risk factors.

Adverse Events

As shown in Table 3, there was no significant difference in the number of subjects in the two groups who died or were hospitalized, although those in the inter-intervention group had fewer hospitalizations and fewer hospital days. The musculoskeletal symptoms of 10 subjects in the intervention group were thought to be probably related to the exercise program. All these events were self-limited.

Cost Effectiveness

The total cost of the intervention, including development, equipment, personnel, travel, and overhead costs, was $136,318, or an average of $891 per subject in the intervention group. The cost per fall prevented ($136,318 divided by 70 [164 falls in the control group - 94 in the intervention group]) was $1,947. The cost for preventing one fall that required medical care was $12,392.

Contrary to the results of earlier intervention trials,^15-18 the targeted-intervention strategy reported here was associated with a reduction in the proportion of subjects who fell and in the incidence of falls. Although the numbers were small, the subjects in the intervention group also reported fewer injuries and fewer episodes of medical care associated with falls. Selecting study subjects at risk for falling, targeting multiple modifiable risk factors, and implementing the interventions ourselves may all have contributed to the effectiveness of the intervention. Although it is not possible to determine whether risk-factor reduction was definitely responsible for the decrease in falls, the reduction in the number of risk factors in a larger proportion of the intervention group than of the control group, the reduction in the effect of treatment when the change in risk factors was added to the model, and the relation between the change in the number of risk factors and the incidence of falls all suggest that risk-factor reduction at least partially explained the decrease in the occurrence of falls.

The greatest difference between the two groups was among subjects with impairments in balance or transfer skills and among those who took four or more prescription medications. The effect on medications in the subjects in the intervention group could have resulted directly from the nurse's counseling the patients and contacting physicians. In addition, several physicians reported a heightened awareness of the relation between medications and falling and of the frequency of postural hypotension. Similar proportions of subjects in the control and intervention groups reported no longer using sedatives at the time of reassessment, perhaps reflecting intermittent use of these medications.

The absence of a greater effect of the interventions on muscle strength could be explained by the fact that manual muscle testing is insensitive to change. Also, because of the priorities assigned to the various risk factors, some subjects with muscle weakness did not receive instructions for a strengthening exercise program. Even for the muscle groups exercised, the intensity or duration of the exercise program may not have been great enough to effect improvement.

As is often observed in clinical trials, a high proportion of control subjects had reductions or improvement in the risk factors⁴². Inherent variability in the risk factors, heightened motivation among subjects participating in a trial, and reassessment selectively biased toward healthier subjects could all explain the improvement in the control subjects. Regardless of the reasons, the higher proportion of subjects in the intervention group who had improvement in blinded assessments suggests that the intervention did have a positive effect on these risk factors.

Several of our methods deserve comment. The use of a matched-block design, rather than purely random sampling, was necessary to prevent the contamination of results that might have occurred if physicians had cared for subjects in both control and intervention groups¹⁹. The groups appeared well matched, and simulation modeling revealed a similar effect of the intervention among all the physicians' patient groups.

As is typical of patients enrolled in HMOs, a high proportion of the subjects were white and had more than a high-school education. We believe the selection of the HMO setting was essential to ensure that the physicians seen by the subjects were an integral part of the medication protocols. The generalizability of the effects of the intervention will need to be determined in other health care settings.

More complete reporting of falls could have occurred in the intervention group than in the control group because of the greater attention received by the subjects in the intervention group, whereas underreporting could have occurred if those subjects had perceived that investigators wanted to see a reduction in falls. Bias alone, however, is unlikely to explain fully the differences between the two groups, since similar trends were seen in terms of falls requiring medical care and falls causing injury, events less susceptible to reporting bias.

This risk-reduction strategy could readily be implemented as part of the clinical care of elderly patients. Third-party reimbursement will be more likely if there is evidence of savings due to treatments that were prevented. For falls requiring medical care, the cost per fall prevented -- $12,392 -- compares favorably with a mean charge of $11,800 per hospitalization for injuries caused by falls among persons 65 years of age or older⁴³. Indeed, a complete analysis of total and fall-related health care costs may show the intervention to result in a net cost savings.

Finally, because many of the risk factors for falling also contribute to immobility and functional decline, the strategy studied here could result in a reduction in the incidence of falls as well as in an improvement in functional independence among elderly patients. This improvement in function is likely, since the intervention not only reduced the number of risk factors and decreased the incidence of falls, but also increased the subjects' confidence in performing daily activities -- an important independent determinant of daily functioning⁴⁴.