A classification and regression tree for predicting recurrent falling among community-dwelling seniors using home-care services.

Approximately 30% of community-dwelling persons, aged 65 or older, fall at least once per year, and about 15% sustain multiple falls. (1-4) Multiple falls are associated with an increased risk of institutionalization and death. (4,5) In addition to injury, recurrent falls can reduce self-confidence, mobility, and social contacts. (6)

Numerous factors might contribute to falls. (7) Some can be corrected and, thus, the event can be avoided. The most efficient interventions are those which target screened fallers with the highest risk of falling again, rather than elderly people identified indiscriminately. (4,7,8)

The increasing number of elderly people is leading to greater demand for home-care services. Preventing falls among community-dwelling seniors using home-care services has become a priority in Quebec. (7,9) Nonetheless, risk factors for falling are overlooked in this specific population. (4,6)

Clinicians are interested in predicting adverse outcomes. The aim of this study has been to develop profiles for predicting the risk of recurrent falling, using a classification and regression tree-based survival analysis.

METHODS

Setting and subjects

The sample studied here was a convenience sample of volunteers recruited between March 2002 and July 2005 among community-living persons, aged 65 years or older, who were receiving public home nursing care, personal care and support services because of a temporary disability or a loss of functional autonomy. (10) People who could speak neither French nor English, those not able to walk more than six metres, and those with reduced communication and cognition according to the Functional Autonomy Measurement System (10) were excluded. All subjects gave informed consent. The study was approved by the authorities of each participating centre. Additional methodological details are provided elsewhere. (11) Of the 959 persons who met the study inclusion criteria and agreed to participate, 868 participants were used in the analyses (Figure 1).

Assessment of falls and predictors

A fall was defined as an event resulting in the subject inadvertently coming to rest on the ground, floor, or other lower level. Excluded were sports-related falls. (1) The outcome was measured by self-report using monthly telephone questionnaire. A falls calendar was previously given to individuals to mark events each time they appear. Recurrent fallers were subjects who had fallen twice within the first six months of follow-up. (3,12)

Potential predictors of recurrent falling and subjects' characteristics were ascertained at baseline at home. Number of falls in the prior three months was categorized as 0, 1, or [greater than or equal to]2. Nutritional screening was performed on a graded 13-point scale to identify individuals at high risk of energy and nutritional intake deficiencies. Pre-established categories were defined as follows: 0-2, 3-5, and 6-13. (13,14) Body weight was self-reported and height was measured. BMI values were defined as [less than or equal to]20, 21-29, and [greater than or equal to]30. Gait and balance were assessed by the Berg scale (15-17) on a 56-point scale ([less than or equal to]30, 31-44, and [greater than or equal to]45), and by the Timed Up & Go test (18,19) which measures the overall time, in seconds, to complete a series of functional tasks ([less than or equal to]20, 21-29, and [greater than or equal to]30). The cutoff values used in the study are those proposed by the developers of each clinical risk assessment tool. Data about the use of benzodiazepines (yes/no) and number of daily consumed prescribed drugs were recorded from the containers. A history of alcohol consumption was obtained according to the Institut de la statistique du Quebec questionnaire. (20,21) Responses were categorized for both drinking in the preceding week (yes/no) and usual drinking during the last six months (nondrinker, <1-3 times a month, 1-6 times a week, every day). Subjects' homes were assessed for 37 environmental hazards using a standardized checklist with unknown validity and reliability. (22-24) Overall and room-specific hazard scores were computed by counting the number of home hazards. Housing types included: single-family house; apartment building; row housing or other unique-entrance dwelling units; private community-based retirement facilities; other housing, including room in shared accommodation. Generally speaking, higher values of the measurements denoted higher risk, except for the Berg score where the opposite was true.

Statistical analyses

Statistical analyses were carried out using SPSS[R] version 15.0. We considered two-tailed p-values less than 0.05 as significant. A two-step analysis was performed to develop risk factor profiles for the prediction of recurrent falling. (25)

We first performed a classification and regression tree (CART) analysis using AnswerTree SPSS[R] version 3.1.1. CART is a technique that recursively splits a parent group into two subgroups (called "nodes") within which covariates are homogenous and between which outcome is distinct. (26,27) The partitioning algorithm started with the covariate and split threshold that best maximized the difference in the outcome between two subgroups. The process was repeated until the first occurrence of either: no covariate significantly dichotomized the nodes further nor did any subgroup reach a minimum specified size (parent node of 25 subjects and child node of 15).

[FIGURE 1 OMITTED]

We next conducted a Kaplan-Meier analysis (28,29) within each group to compare their risk of falling over time. The survival time was defined as the time to the second fall where the falls occurred less than six months apart. Data were censored upon reaching a specified time point earlier, withdrawal for any reason, or end of follow-up period. The log-rank (Mantel-Cox) test identified differences in the cumulative survival curves for all possible pairs of terminal nodes identified. The Bonferroni-Holm correction of alpha value was used for multiple comparisons. (30)

RESULTS

Altogether 99 of 868 participants reported two falls within six months of entry to the study. Thus, the incidence for recurrent falling was 11.4%. Of the 769 non-recurrent fallers, 151 (19.6%) reported one fall, while the others did not declare any event. Study subjects were mainly women (77.2%), of whom 76.2% were 75 years or older. Table 1 shows the baseline characteristics of non- and recurrent fallers. The recurrent fallers included significantly more males, had a lower performance of gait and balance, and experienced falls more often than non-recurrent fallers in the three months prior to study entry.

Figure 2 shows the classification tree results for predicting recurrent fallers. The root node (no 0) comprising the entire sample (n = 868) was first separated into nodes no 1 and no 2, involving the history of falls. The 29.8% of the participants from node no 2, falling at least two times in the three months prior to study entry, became recurrent fallers during the following six months, whereas the 8.1% of participants from node no 1, falling fewer than two times, became recurrent fallers. The analysis identified four end nodes for participants who fell at least two times in the three prior months. Among them, subjects who scored [less than or equal to]30 on the Berg test and who reported consuming alcohol within the six months prior to baseline formed the end node no 8 with the greatest relative risk (RR = 5.1) of becoming recurrent fallers (Figure 2 and Table 2). Conversely, those who sustained fewer than two falls prior to baseline remained undivided (node no 1) and formed the most favourable group (RR = 0.7). Table 2 summarizes each terminal node. Three profiles (node nos 8, 5 and 7) leave people at a significantly (p[less than or equal to]0.01) higher than average risk of becoming recurrent fallers during the follow-up (RR = 5.1, 2.5 and 2.1).

Survival curves for each of the terminal nodes are presented in Figure 3. Subjects in node no 8, who fell earlier (average survival time = 124 days), were those who fell more often (RR = 5.1). Conversely, those in node no 1, who fell later (average survival time = 176 days), were those who fell less often (RR = 0.7). The probability of not being a recurrent faller over the six-month follow-up is equivalent to 100 minus the incidence of recurrent falling. Statistical difference in the cumulative survival curves of each node are given in Table 3.

DISCUSSION

The CART and survival analyses divided the population into five specific combinations of predictors, and characterized them by an estimated risk of becoming a recurrent faller and of length of time before becoming a recurrent faller.

The methodology identified three profiles of higher short-term risk of recurrent falling among the community-dwelling elderly who use home-care services. A history of falls in the three months prior to the initial interview emerged as a predictor of recurrent falling; results seemed to indicate that a recurrent faller was likely to remain so. Also at a high risk of becoming recurrent fallers were participants with [greater than or equal to]2 prior falls and a score of [less than or equal to]30 on the Berg balance scale (node no 4)--particularly those who drank alcohol in the six months preceding their examination (node no 8)--as well as participants with as many prior falls but with a score higher than 30 on the Berg scale, who lived in a private residential facility (node no 5).

[FIGURE 2 OMITTED]

Leclerc et al. (31) have previously compared different statistical methods to identify predictors of falls among community-dwelling seniors who use home-care services. They have shown that a history of falling, the Berg balance score, and residential facility housing each appear to be a predictor of falls, whatever the outcomes (number of falls, time to first fall, and time to each recurrent fall). Our results concur with authors who consider a history of falls in the previous year as high-risk criteria for falling and for eligibility in intervention programs. (7) Nevertheless, not all subjects with a history of falls run a greater risk of becoming recurrent fallers. Notably, elderly people in node no 1 were significantly less likely than on average, even if they had already fallen once.

[FIGURE 3 OMITTED]

The first and only study applying a tree-based methodology for the prediction of falls was conducted by Stel et al. (3) Although they did not use the same subset of predictors as we did, the history of falls in the previous year clearly emerged as the prime predictor of recurrent falls. They also used a similar approach, the tree-structured survival analysis. We preferred not to use it for the following reasons. It was developed for running on S-plus 3 by two programmers external to the statistical firm. (32,33) The program has not been updated and TSSA is no longer supported by either the current versions of S-plus or the firm. We also noted that the algorithm implemented in the program could continue to divide a parent node, even in the absence of statistical significance of a split.

The tree-based methodology provides a number of advantages over linear, logistic, and Cox regression models. First, it is a nonparametric and non-linear technique. It does not require any a priori distributional assumption and knowledge about the underlying relationships between the predictors and the dependent variable. (3,27,33,34) Hence, the method is useful in situations where there are interactions among variables: the cases are partitioned and each group is analyzed separately. Second, CART allows the construction of directly applicable fall risk profiles. Contrary to regression analyses where all predictors must be measured in order to identify the risk of falls, few predictors should be required to recognize the risk profile of a new case. (3)

A number of cautions arise from our study. First, monitoring falls in cohort studies relies on some form of self-report, typically done by having participants recall whether or not they have fallen over a designated time period. Active registration by the participant and a monthly phone call delivered by a caregiver reduce the problem of recall over long periods, but require that participants remember and make the effort to mark the event after having experienced a fall. Further, regardless of the method of self-reporting, seniors may be reluctant to admit they have fallen. These limitations around self-report have led to the conclusion that reliance on self-report likely produces an under-reporting. (35) Second, the findings with CART do not necessarily imply cause and effect relationships, but simple statistical associations. Third, we should remember that any data-driven clustering results must be validated by using a separate sample. Fourth, the question of what constitutes an appropriately sized tree remains unresolved. We selected stop-splitting criteria in order to attain a balance between the purity of the end-branching nodes and a reasonable number of subjects for clinical significance. Finally, biases may have occurred because of differences between the retained participants and the individuals lost to follow-up. Males, high nutritional risk (values [greater than or equal to]6), lower Berg score, as well as impaired balance (Berg [less than or equal to]30) and mobility (TUG [greater than or equal to]30) at baseline, were more likely to be lost to follow-up. This would lead to an underestimation of the effects.

Our results tend to support the theory that multiple falls may have more intrinsic causes than a single fall, (3,36) especially if the variables of "history of falling" and "living in a private residential facility" act as a surrogate measure of various chronic conditions and poorer functional autonomy.

Acknowledgements: The authors wish to thank all older clients and health care workers from the community health and social service centres in Lanaudiere for their participation in the study. We also acknowledge the contribution of Genevieve Marquis for the data entry, Josee Payette for the data processing, and Bruce Charles Bezeau for the revision of the manuscript. The research was sponsored by the Agence de la sante et des services sociaux de Lanaudiere and the Groupe de recherche interdisciplinaire en sante of the Universite de Montreal.

Received: December 28, 2007

Revisions requested: March 13, 2008 and September 10, 2008

Revised mss: July 1, 2008 and February 22, 2009

Accepted: March 9, 2009

REFERENCES

(1.) O'Loughlin JL, Robitaille Y, Boivin JF, Suissa S. Incidence of and risk factors for falls and injurious falls among the community-dwelling elderly. Am J Epidemiol 1993;137(3):342-54.

(2.) Hill K, Schwarz J, Flicker L, Carroll S. Falls among healthy, communitydwelling, older women: A prospective study of frequency, circumstances, consequences and prediction accuracy. Aust N Z J Public Health 1999;23(1):41-48.

(3.) Stel VS, Pluijm SM, Deeg DJ, Smit JH, Bouter LM, Lips P. A classification tree for predicting recurrent falling in community-dwelling older persons. J Am Geriatr Soc 2003;51(10):1356-64.

(4.) Fletcher PC, Hirdes JP. Risk factors for falling among community-based seniors using home care services. J Gerontol A Biol Sci Med Sci 2002;57(8):M504-10.

(5.) Donald IP, Bulpitt CJ. The prognosis of falls in elderly people living at home. Age Ageing 1999;28(2):121-25.

(6.) Fletcher PC, Hirdes JP. Restriction in activity associated with fear of falling among community-based seniors using home care services. Age Ageing 2004;33(3):273-79.

(7.) Ministere de la Sante et des Services sociaux du Quebec. La prevention des chutes dans un continuum de services pour les aines vivant a domicile, Cadre de reference, Quebec, Direction generale de la sante publique, 2004; 61 p.

(8.) Gardner MM, Robertson MC, Campbell AJ. Exercise in preventing falls and fall related injuries in older people: A review of randomised controlled trials. Br J Sports Med 2000;34(1):7-17.

(9.) Ministere de la Sante et des Services sociaux du Quebec. Programme national de sante publique 2003-2012, Quebec, Direction generale de la sante publique, 2003; 133 p.

(10.) Tousignant M, Dubuc N, Hebert R, Coulombe C. Home-care programmes for older adults with disabilities in Canada: How can we assess the adequacy of services provided compared with the needs of users? Health Soc Care Community 2007;15(1):1-7.

(11.) Begin C. Projet-pilote regional de prevention des chutes a domicile chez les personnes agees, Devis d'implantation dans les CLSC, Saint-Charles-Borromee, Service de prevention et de promotion, Direction de sante publique, Regie regionale de la sante et des services sociaux de Lanaudiere, 2002; 120 p.

(12.) Pluijm SM, Smit JH, Tromp EA, Stel VS, Deeg DJ, Bouter LM, Lips P. A risk profile for identifying community-dwelling elderly with a high risk of recurrent falling: Results of a 3-year prospective study. Osteoporos Int 2006;17(3):417-25.

(13.) Laporte M, Villalon L, Payette H. Simple nutrition screening tools for healthcare facilities: Development and validity assessment. Can J Diet Pract Res 2001;62(1):26-34.

(14.) Laporte M, Villalon L, Thibodeau J, Payette H. Validity and reliability of simple nutrition screening tools adapted to the elderly population in healthcare facilities. J Nutr Health Aging 2001;5(4):292-94.

(15.) Berg KO, Maki BE, Williams JI, Holliday PJ, Wood-Dauphinee SL. Clinical and laboratory measures of postural balance in an elderly population. Arch Phys Med Rehabil 1992;73(11):1073-80.

(16.) Berg KO, Wood-Dauphinee SL, Williams JI, Maki B. Measuring balance in the elderly: Validation of an instrument. Can J Public Health 1992;83(Suppl. 2):S7-S11.

(17.) Berg K, Wood-Dauphinee S, Williams JI. The Balance Scale: Reliability assessment with elderly residents and patients with an acute stroke. Scand J Rehabil Med 1995;27(1):27-36.

(18.) Podsiadlo D, Richardson S. The timed "Up & Go": A test of basic functional mobility for frail elderly persons. J Am Geriatr Soc 1991;39(2):142-48.

(19.) Lin MR, Hwang HF, Hu MH, Wu HD, Wang YW, Huang FC. Psychometric comparisons of the timed up and go, one-leg stand, functional reach, and Tinetti balance measures in community-dwelling older people. J Am Geriatr Soc 2004;52(8):1343-48.

(20.) Chevalier S, Lemoine O. Consommation d'alcool, in Enquete sociale et de sante 1998 (Collection La sante et le bien-etre), Quebec: Institut de la statistique du Quebec, 2000;117-33.

(21.) Institut de la statistique du Quebec. Questionnaire autoadministre (QAA) VI--L'alcool, in Enquete sociale et de sante 1998 (Collection La sante et le bienetre), Quebec: Institut de la statistique du Quebec, 2000, p. 15-17.

(22.) Gill TM, Williams CS, Robison JT, Tinetti ME. A population-based study of environmental hazards in the homes of older persons. Am J Public Health 1999;89(4):553-56.

(23.) Gill TM, Williams CS, Tinetti ME. Environmental hazards and the risk of nonsyncopal falls in the homes of community-living older persons. Med Care 2000;38(12):1174-83.

(24.) Reseau francophone de prevention des traumatismes et de promotion de la securite. Referentiel de bonnes pratiques. Prevention des chutes chez les personnes agees a domicile, Paris: editions INPES, 2005; 156 p.

(25.) Fan Z, Kabrick JM, Shifley SR. Classification and regression tree based survival analysis in oak-dominated forests of Missouri's Ozark highlands. Can J Forest Res 2006;36(7):1740-48.

(26.) Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and Regression Trees. Monterey, CA: Wadsworth and Brooks/Cole, 1984; 358 p.

(27.) Madigan EA, Curet OL. Madigan EA, Curet OL. A data mining approach in home healthcare: Outcomes and service use. BMC Health Serv Res 2006;6:18 doi:10.1186/1472-6963-6-18. Available online at: www.biomedcentral.com/1472-6963/6/18 (Accessed June 5, 2009).

(28.) Hosmer DW, Lemeshow S. Applied Survival Analysis: Regression Modeling of Time to Event Data. New York, NY: John Wiley and Sons Inc., 1999; 386 p.

(29.) Allison PD. Survival Analysis Using SAS: A Practical Guide. Cary, NC: SAS Institute Inc., 1995; 304 p.

(30.) Holm S. A simple sequentially rejective multiple test procedure. Scand J Statist 1979;6:65-70.

(31.) Leclerc BS, Begin C, Cadieux E, Goulet L, Leduc N, Kergoat MJ, Lebel P. Risk factors for falling among community-dwelling seniors using home-care services: An extended hazards model with time-dependent covariates and multiple events. Chron Dis Can 2008;28(4):111-20.

(32.) Segal MR. Regression trees for censored data. Biometrics 1988;44:35-47.

(33.) Segal MR. Features of tree-structured survival analysis. Epidemiology 1997;8:344-46.

(34.) Clark TG, Bradburn MJ, Love SB, Altman DG. Survival Analysis Part IV: Further concepts and methods in survival analysis. Br J Cancer 2003;89(5):78186.

(35.) Ganz DA, Higashi T, Rubenstein LZ. Monitoring falls in cohort studies of community-dwelling older people: Effect of the recall interval. J Am Geriatr Soc 2005;53(12):2190-94.

(36.) Nevitt MC, Cummings SR, Kidd S, Black D. Risk factors for recurrent nonsyncopal falls. A prospective study. JAMA 1989;261(18):2663-68.

Bernard S. Leclerc, MSc, [1] * Claude Begin, MSc, [2] Elizabeth Cadieux, MSc, [1] Lise Goulet, MD, PhD, [3] Jean-Francois Allaire, MSc, [4] Julie Meloche, MSc, [4] Nicole Leduc, PhD, [3] Marie-Jeanne Kergoat, MD, CCFP, FCFP, CSPQ [5]

Author Affiliations

[1.] Service de surveillance, recherche et evaluation, Direction de sante publique et d'evaluation, Agence de la sante et des services sociaux de Lanaudiere, Joliette, QC (at the time of the study)

[2.] Service de prevention et de promotion, Direction de sante publique et d'evaluation, Agence de la sante et des services sociaux de Lanaudiere, Joliette, QC

[3.] Groupe de recherche interdisciplinaire en sante, Universite de Montreal, Montreal, QC

[4.] The Statistics Consulting Group, Institut Philippe-Pinel de Montreal, Montreal, QC

[5.] Research Centre, Institut universitaire de geriatrie de Montreal, Montreal, QC

* This research is part of Leclerc's PhD thesis in Public Health and Epidemiology, realized under the supervision of Professors Lise Goulet and Nicole Leduc, respectively from the Departement de medecine sociale et preventive and the Departement d'administration de la sante, Faculte de medecine, Universite de Montreal, Montreal, QC, Canada.

Correspondence: Bernard-Simon Leclerc, Direction Developpement des individus et des communautes, Institut national de sante publique du Quebec, 190, boul. Cremazie est, Montreal, QC H2P 1E2, Tel: [phone], ext. 3530, Fax: [phone], E-mail: [e-mail].

Table 1. Baseline Characteristics of the Community-dwelling
Elderly, According to Their Status

Risk Factor                                      Non-recurrent
                                                 Fallers
                                                 (n = 769)

Socio-demographic variables
  Age (yrs), [bar.x] [+ or -] SD                 79.5 [+ or -] 6.6
  [greater than or equal to]75 yrs, %            75.9
  Male, %                                        22
Type of residence
  House/single-family home, %                    57.1
  Residential facility, %                        12.2
Home hazards, [bar.x] [+ or -] SD                3.2 [+ or -] 2.3
Body composition
  BMI (kg/m2), [bar.x] [+ or -] SD               28.1 [+ or -] 6.9
  Underweight (BMI [less than or equal           10.1
    to]20), %
  Obesity (BMI [greater than or equal            34.7
    to]30), %
Nutrition
  Screening score, [bar.x] [+ or -] SD           3.7 [+ or -] 1.9
  High nutritional risk (values [greater         15.6
    than or equal to]6), %
Gait, balance and mobility
  Berg balance score, [bar.x] [+ or -] SD        43.9 [+ or -] 8.8
  Impaired balance (Berg [less than or           8.6
    equal to]30), %
  Timed Up & Go score (sec), [bar.x]             23.5 [+ or -] 16.7
    [+ or -] SD
  Impaired mobility (TUG [greater than or        17.9
    equal to]30), %
Medication use
  Distinct prescribed drugs daily, [bar.x]       8.7 [+ or -] 4.2
    [+ or -] SD
  [greater than or equal to]4 prescribed         87.1
    drugs per day, %
  Benzodiazepine use, %                          47.2
Alcohol use
  Consumption in past 7 days, %                  20.2
  Consumption in past 6 months, %                46.3
History of falls, in past 3 months, %            35.5

Risk Factor                                      Recurrent
                                                 Fallers
                                                 (n = 99)

Socio-demographic variables
  Age (yrs), [bar.x] [+ or -] SD                 79.0 [+ or -] 6.9
  [greater than or equal to]75 yrs, %            77.8
  Male, %                                        31.3 *
Type of residence
  House/single-family home, %                    54.5
  Residential facility, %                        18.2
Home hazards, [bar.x] [+ or -] SD                3.5 [+ or -] 2.6
Body composition
  BMI (kg/m2), [bar.x] [+ or -] SD               27.4 [+ or -] 5.3
  Underweight (BMI [less than or equal           8.1
    to]20), %
  Obesity (BMI [greater than or equal            30.3
    to]30), %
Nutrition
  Screening score, [bar.x] [+ or -] SD           3.8 [+ or -] 2.0
  High nutritional risk (values [greater         20.2
    than or equal to]6), %
Gait, balance and mobility
  Berg balance score, [bar.x] [+ or -] SD        39.5 [+ or -] 8.5 ***
  Impaired balance (Berg [less than or           15.2*
    equal to]30), %
  Timed Up & Go score (sec), [bar.x]             27.6 [+ or -] 17.2 *
    [+ or -] SD
  Impaired mobility (TUG [greater than or        25.3
    equal to]30), %
Medication use
  Distinct prescribed drugs daily, [bar.x]       9.4 [+ or -] 4.1
    [+ or -] SD
  [greater than or equal to]4 prescribed         91.9
    drugs per day, %
  Benzodiazepine use, %                          50.5
Alcohol use
  Consumption in past 7 days, %                  21.2
  Consumption in past 6 months, %                52.5
History of falls, in past 3 months, %            65.7 ***

* p[less than or equal to]0.05; ** p[less than or equal to]0.01;
*** p[less than or equal to]0.001

Table 2. Summary of the Tree for Predicting Recurrent Fallers at
Six-month Follow-up among Community-dwelling Seniors Using Home-care
Services, in Descendant Order According to the Relative Risk

Node no. and Profile          Number of   Incidence of   Relative
                              Subjects    Recurrent      Risk
                                          Fallers (%)

#8: [greater than or          33          57.6 *         5.1
  equal to]2 falls in
  prior three months,
  [greater than or equal
  to]30 on Berg balance
  scale, alcohol intake
  in prior six months
#5: [greater than or          21          28.6 *         2.5
  equal to]2 falls in
  prior three months,
  >30 on Berg balance
  scale, living in
  private residential
  facility
#7: [greater than or          45          24.4 *         2.1
  equal to]2 falls in
  prior three months,
  [greater than or equal
  to]30 on Berg balance
  scale, no prior alcohol
  intake
#6: [greater than or
  equal to]2 falls in
  prior three months,
  >30 on Berg balance
  scale, living in single
  family house or other
  types of personal
  housing                     32          9.4            0.8
#1: <2 falls in prior         737         8.1 *          0.7
  three months
Total                         868         11.4           1.0

* The incidence of recurrent fallers in node differs significantly
from the total sample incidence at the 0.05 level, after
Bonferroni-Holm adjustment for multiple comparisons.

Table 3. Pairwise Comparisons of Survival Curves for
Predicting Recurrent Fallers at Six-month Follow-up
among Community-dwelling Seniors Using Homecare
Services, According to the Log-rank
(Mantel-Cox) Test

Node no.   1         5       6         7         8

1          1.000
5          0.000 *   1.000
6          0.780     0.071   1.000
7          0.000 *   0.712   0.094     1.000
8          0.000 *   0.045   0.000 *   0.003 *   1.000

* Significantly different at the 0.05 level (two-tailed test),
after Bonferroni-Holm adjustment for multiple comparisons.