Advertisement

Clinical Prediction Rule for Declines in Activities of Daily Living at 6 Months After Surgery for Hip Fracture Repair

Open AccessPublished:August 13, 2016DOI:https://doi.org/10.1016/j.apmr.2016.07.016

      Abstract

      Objectives

      To develop and assess a clinical prediction rule (CPR) to predict declines in activities of daily living (ADL) at 6 months after surgery for hip fracture repair.

      Design

      Prospective, cohort study.

      Setting

      From hospital to home.

      Participants

      Patients (N=104) with hip fractures after surgery.

      Interventions

      Not applicable.

      Main Outcome Measure

      ADL were assessed using the Barthel Index at 6 months after surgery.

      Results

      At 6 months after surgery, 86 patients (82.6%) were known to be alive, 1 patient (1.0%) had died, and 17 (16.3%) were lost to follow-up. Thirty-two patients (37.2%) did not recover their ADL at 6 months after surgery to levels before fracture. The classification and regression trees methodology was used to develop 2 models to predict a decline in ADL: (1) model 1 included age, type of fracture, and care level before fracture (sensitivity=75.0%, specificity=81.5%, positive predictive value=70.6%, positive likelihood ratio=4.050); and (2) model 2 included the degree of independence 2 weeks postsurgery for ADL chair transfer, ADL ambulation, and age (sensitivity=65.6%, specificity=87.0%, positive predictive value=75.0%, positive likelihood ratio=5.063). The areas under the receiver operating characteristic curves of both CPR models were .825 (95% confidential interval, .728–.923) and .790 (95% confidence interval, .683–.897), respectively.

      Conclusions

      CPRs with moderate accuracy were developed to predict declines in ADL at 6 months after surgery for hip fracture repair.

      Keywords

      List of abbreviations:

      ADL (activities of daily living), AUROC (area under the receiver operating characteristic curve), CART (classification and regression tree), CPR (clinical prediction rule), CS-30 (chair-stand-30), PLR (positive likelihood ratio), PPV (positive predictive value)
      The incidence of hip fracture has continued to increase among the ageing population.
      • Johnell O.
      • Kanis J.
      Epidemiology of osteoporotic fractures.
      • Sambrook P.
      • Cooper C.
      Osteoporosis.
      Assuming that the age-related incidence will increase by only 1% per year, the number of hip fractures globally is expected to reach 8.2 million in 2050,
      • Sambrook P.
      • Cooper C.
      Osteoporosis.
      with approximately half of the patients unable to regain the ability to live independently.
      • Morrison R.S.
      • Chassin M.R.
      • Siu A.L.
      The medical consultant's role in caring for patients with hip fracture.
      In 1997, the direct and indirect annual costs of hip fracture treatment were estimated at $131.5 billion globally.
      • Johnell O.
      The socioeconomic burden of fractures: today and in the 21st century.
      To prevent increased medical costs, health care providers should make efforts to identify persons at risk for declines in their activities of daily living (ADL) after discharge.
      The following factors before or at the time of fracture have been associated with a decline in ADL at 6 months, 1 year, or 2 years after surgery: age
      • Penrod J.D.
      • Litke A.
      • Hawkes W.G.
      • et al.
      Heterogeneity in hip fracture patients: age, functional status, and comorbidity.
      • Haentjens P.
      • Autier P.
      • Barette M.
      • Boonen S.
      Belgian Hip Fracture Study Group
      Predictors of functional outcome following intracapsular hip fracture in elderly women. A one-year prospective cohort study.
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      • Eastwood E.A.
      • Magaziner J.
      • Wang J.
      • et al.
      Patients with hip fracture: subgroups and their outcomes.
      ; ADL
      • Penrod J.D.
      • Litke A.
      • Hawkes W.G.
      • et al.
      Heterogeneity in hip fracture patients: age, functional status, and comorbidity.
      • Eastwood E.A.
      • Magaziner J.
      • Wang J.
      • et al.
      Patients with hip fracture: subgroups and their outcomes.
      • Kagaya H.
      • Takahashi H.
      • Sugawara K.
      • Dobashi M.
      • Kiyokawa N.
      • Ebina H.
      Predicting outcomes after hip fracture repair.
      • Alegre-Lopez J.
      • Cordero-Guevara J.
      • Alonso-Valdivielso J.L.
      • Fernandez-Melon J.
      Factors associated with mortality and functional disability after hip fracture: an inception cohort study.
      ; living in an institution at the time of injury
      • Haentjens P.
      • Autier P.
      • Barette M.
      • Boonen S.
      Belgian Hip Fracture Study Group
      Predictors of functional outcome following intracapsular hip fracture in elderly women. A one-year prospective cohort study.
      • Eastwood E.A.
      • Magaziner J.
      • Wang J.
      • et al.
      Patients with hip fracture: subgroups and their outcomes.
      ; the ability to shop for self or the degree of independent walking
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      • Ingemarsson A.H.
      • Frandin K.
      • Mellstrom D.
      • Moller M.
      Walking ability and activity level after hip fracture in the elderly—a follow-up.
      ; type of fracture
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      ; concomitant diseases
      • Shyu Y.I.
      • Chen M.C.
      • Liang J.
      • Wu C.C.
      • Su J.Y.
      Predictors of functional recovery for hip fractured elders during 12 months following hospital discharge: a prospective study on a Taiwanese sample.
      • Kim S.M.
      • Moon Y.W.
      • Lim S.J.
      • et al.
      Prediction of survival, second fracture, and functional recovery following the first hip fracture surgery in elderly patients.
      ; and cognitive impairment.
      • Penrod J.D.
      • Litke A.
      • Hawkes W.G.
      • et al.
      Heterogeneity in hip fracture patients: age, functional status, and comorbidity.
      • Kim S.M.
      • Moon Y.W.
      • Lim S.J.
      • et al.
      Prediction of survival, second fracture, and functional recovery following the first hip fracture surgery in elderly patients.
      Predictors after fracture include a shorter hospital stay,
      • Shyu Y.I.
      • Chen M.C.
      • Liang J.
      • Wu C.C.
      • Su J.Y.
      Predictors of functional recovery for hip fractured elders during 12 months following hospital discharge: a prospective study on a Taiwanese sample.
      poor functional status,
      • Haentjens P.
      • Autier P.
      • Barette M.
      • Boonen S.
      Belgian Hip Fracture Study Group
      Predictors of functional outcome following intracapsular hip fracture in elderly women. A one-year prospective cohort study.
      • Ingemarsson A.H.
      • Frandin K.
      • Mellstrom D.
      • Moller M.
      Walking ability and activity level after hip fracture in the elderly—a follow-up.
      poor mental status,
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      • Alegre-Lopez J.
      • Cordero-Guevara J.
      • Alonso-Valdivielso J.L.
      • Fernandez-Melon J.
      Factors associated with mortality and functional disability after hip fracture: an inception cohort study.
      poor physical health,
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      need for a caregiver at discharge,
      • Shyu Y.I.
      • Chen M.C.
      • Wu C.C.
      • Cheng H.S.
      Family caregivers' needs predict functional recovery of older care recipients after hip fracture.
      emotional support at 1 month after discharge,
      • Shyu Y.I.
      • Tang W.R.
      • Tsai W.C.
      • Liang J.
      • Chen M.C.
      Emotional support levels can predict physical functioning and health related quality of life among elderly Taiwanese with hip fractures.
      impaired cognitive functioning,
      • Oude Voshaar R.C.
      • Banerjee S.
      • Horan M.
      • et al.
      Fear of falling more important than pain and depression for functional recovery after surgery for hip fracture in older people.
      and fear of falling
      • Oude Voshaar R.C.
      • Banerjee S.
      • Horan M.
      • et al.
      Fear of falling more important than pain and depression for functional recovery after surgery for hip fracture in older people.
      assessed 6 weeks after surgery. Although potential predictors before or after fracture have been widely investigated, the predictive accuracies of these factors to predict a decline in ADL after surgery remain unknown. Hence, individual predictors and in combination must be evaluated.
      Various clinical prediction rules (CPRs) have been developed to identify the best combination of medical signs, symptoms, and other findings to predict the probability of a specific disease or treatment outcome.
      • McGinn T.G.
      • Guyatt G.H.
      • Wyer P.C.
      • Naylor C.D.
      • Stiell I.G.
      • Richardson W.S.
      Users' guides to the medical literature: XXII: how to use articles about clinical decision rules. Evidence-Based Medicine Working Group.
      CPRs provide health care providers with an evidence-based tool to assist in patient management. For example, the Harris hip score
      • Harris W.H.
      Traumatic arthritis of the hip after dislocation and acetabular fractures: treatment by mold arthroplasty. An end-result study using a new method of result evaluation.
      was developed to assess the results of hip surgery or hip replacement. In addition, some CPRs have been developed for use in orthopedics. However, there is no CPR to predict the prognosis of ADL after surgery for hip fracture repair. Therefore, the purpose of this study was to develop and assess the accuracy of a CPR for the decline in ADL 6 months after surgery for hip fracture repair.

      Methods

      Study design

      This prospective, cohort study of patients with hip fractures after surgery was approved by the ethics committees of the 2 hospitals where the participants were recruited (Chugoku Rousai Hospital [reference no. 2014-04] and Saiseikai Kure Hospital [reference no. 110], Kure, Japan) and the ethics committee of the university where the first author was affiliated (Hiroshima International University, Higashihiroshima, Japan [reference no. 14-118]). All study participants provided written informed consent.

      Setting

      The study setting was the hospital and the patients' homes. Recruitment, follow-up, and data collection were performed between April 2013 and September 2015. The staff of the rehabilitation departments of the 2 hospitals recruited potential participants.

      Participants

      Participants were included in the study if they (1) had sustained a fracture of their proximal femur; (2) had undergone a surgical procedure such as femoral head replacement; (3) were predominantly living independently indoors (degree of independent daily living for the disabled elderly of J or A; appendix 1); and (4) had some life-disturbing symptoms daily (ie, behavioral and communication problems) but could independently lead their daily life if watched by someone (independent daily living score for the elderly with dementia of I or II; appendix 2). Because cognitive impairment is another problem associated with locomotive impairment, we did not include elderly patients who had serious cognitive impairment before the fracture. The exclusion criteria were as follows: (1) the existence of hip fracture on the opposite side, and (2) multiple fractures. We prospectively collected patient information from 1 month before the fracture to 6 months after surgery. A total of 104 adults (25 men, 79 women; mean age ± SD, 80.8±10.5y) participated in this study.

      Variables

      All potential predictors investigated were selected from the results of previous studies, which included prerehabilitation baseline data (age, sex, gait method, place of residence, ADL level, nursing care level, type of fracture, surgical method). Prerehabilitation data were collected during patient interviews at the start of rehabilitation by physical therapists (table 1). The type of fracture and surgical method were confirmed from medical records or by discussion with a physician.
      Table 1Data set for developing CPRs
      TimeVariable NameVariable TypeCategory
      Prerehabilitation (baseline)AgeContinuous
      SexCategoricalMale, female
      Gait method before fractureCategoricalIndependent gait/Crutch walking/Gait using a walker/Walking while holding on to something (such as a wall or table)
      Dwelling place before fractureCategoricalHome/institution
      ADL before fractureFeedingOrdinal0=unable; 5=needs help cutting, spreading butter, etc, or requires modified diet; 10=independent
      Chair transferOrdinal0=unable, no sitting balance; 5=major help (1 or 2 people, physical), can sit; 10=minor help (verbal or physical); 15=independent
      GroomingOrdinal0=needs help with personal care; 5=independent face/hair/teeth/shaving (implements provided)
      ToiletingOrdinal0=dependent; 5=needs some help, but can do something alone; 10=independent (on and off, dressing, wiping)
      BathingOrdinal0=dependent; 5=independent (or in shower)
      AmbulationOrdinal0=immobile or <50yd; 5=wheelchair independent, including corners, >50yd; 10=walks with help of 1 person (verbal or physical) >50yd; 15=independent (but may use any aid; eg, stick) >50yd
      Stair climbingOrdinal0=unable; 5=needs help (verbal, physical, carrying aid); 10=independent
      DressingOrdinal0=dependent; 5=needs help but can do about half unaided; 10=independent (including buttons, zips, laces, etc)
      Bowel controlOrdinal0=incontinent (or needs to be given enemas); 5=occasional accident; 10=continent
      Bladder controlOrdinal0=incontinent, or catheterized and unable to manage alone; 5=occasional accident; 10=continent
      Nursing care level before fractureOrdinal0=none; 1=needed support level 1; 2=needed support level 2; 3=care level 1; 4=care level 2; 5=care level 3; 6=care level 4; 7=care level 5
      Type of fractureCategoricalFemoral neck/trochanteric/subtrochanteric
      Surgical formCategoricalFemoral head replacement/gamma nail or alexa nail/compression hip screw/osteosynthesis with plate
      Second week after surgeryADLOrdinal
      PainContinuous
      CS-30Continuous
      ADL disabilities, chair-stand-30 (CS-30) test results, and pain intensity at 2 weeks after surgery were included as predictors. ADL disabilities during hospitalization were assessed, since such functional limitations during hospitalization were found to be strong predictors of mortality
      • McGinn T.G.
      • Guyatt G.H.
      • Wyer P.C.
      • Naylor C.D.
      • Stiell I.G.
      • Richardson W.S.
      Users' guides to the medical literature: XXII: how to use articles about clinical decision rules. Evidence-Based Medicine Working Group.
      • Harris W.H.
      Traumatic arthritis of the hip after dislocation and acetabular fractures: treatment by mold arthroplasty. An end-result study using a new method of result evaluation.
      • Mahoney F.I.
      • Barthel D.W.
      Functional evaluation: the Barthel Index.
      • Jones C.J.
      • Rikli R.E.
      • Beam W.C.
      A 30-s chair-stand test as a measure of lower body strength in community-residing older adults.
      • Della Pietra G.L.
      • Savio K.
      • Oddone E.
      • Reggiani M.
      • Monaco F.
      • Leone M.A.
      Validity and reliability of the Barthel Index administered by telephone.
      • Breiman L.
      • Friedman J.
      • Olshen R.
      • Stone C.
      Classification and regression tree analysis.
      • Aguiar F.S.
      • Almeida L.L.
      • Ruffino-Netto A.
      • Kritski A.L.
      • Mello F.C.
      • Werneck G.L.
      Classification and regression tree (CART) model to predict pulmonary tuberculosis in hospitalized patients.
      and long-term care placement.
      • Westreich D.
      • Lessler J.
      • Funk M.J.
      Propensity score estimation: neural networks, support vector machines, decision trees (CART), and meta-classifiers as alternatives to logistic regression.
      • Takahashi O.
      • Cook E.F.
      • Nakamura T.
      • Saito J.
      • Ikawa F.
      • Fukui T.
      Risk stratification for in-hospital mortality in spontaneous intracerebral haemorrhage: a classification and regression tree analysis.
      • Swets J.A.
      Measuring the accuracy of diagnostic systems.
      • Greiner M.
      • Pfeiffer D.
      • Smith R.D.
      Principles and practical application of the receiver-operating characteristic analysis for diagnostic tests.
      The Barthel Index
      • Mahoney F.I.
      • Barthel D.W.
      Functional evaluation: the Barthel Index.
      is an ordinal scale used to measure performance in ADL (ie, feeding oneself, bathing, dressing, grooming, and the ability to move) on a scale of 0 to 100 (0, very dependent; 100, independent). The scoring was based on objective assessments during hospitalization. The CS-30 test was used to assess functional lower extremity strength.
      • Jones C.J.
      • Rikli R.E.
      • Beam W.C.
      A 30-s chair-stand test as a measure of lower body strength in community-residing older adults.
      The 30-second chair test is administered using a folding chair without arms. At the signal “go,” the participant rose to a full stand (body erect and straight) followed by returning to the initial seated position. Participants were encouraged to complete as many full stands as possible within 30 seconds and were instructed to completely sit between each stand. The scoring was based on the total number of stands within 30 seconds. Excellent test-retest and interrater reliability, and criterion validity of the CS-30 test have been reported.
      • Jones C.J.
      • Rikli R.E.
      • Beam W.C.
      A 30-s chair-stand test as a measure of lower body strength in community-residing older adults.
      Pain was assessed during the CS-30 test using a 100-mm visual analog scale. Data were collected at 2 weeks after surgery by the same therapist (see table 1).
      Patients were followed up during the 6 months after surgery, and telephonic interviews were conducted to assess ADL disability using the Barthel Index, which is reportedly quite reliable.
      • Della Pietra G.L.
      • Savio K.
      • Oddone E.
      • Reggiani M.
      • Monaco F.
      • Leone M.A.
      Validity and reliability of the Barthel Index administered by telephone.
      Interviewers were blinded to patient characteristics at prefracture or during hospitalization. A decline in ADL was defined as a decrease at 6 months after surgery from that before the fracture, scored using the Barthel Index.

      Statistical analysis

      The primary outcome was the Barthel Index total score at 6 months after surgery. In this study, the classification and regression tree (CART) methodology was used.
      • Breiman L.
      • Friedman J.
      • Olshen R.
      • Stone C.
      Classification and regression tree analysis.
      Binary trees are used to recursively split predictor variables with yes/no questions about each variable. Any type of statistical distributions can be handled, and it is not limited to linear relationships between outcome and predictor variables. These algorithms have been used to develop prediction models in various fields.
      • Aguiar F.S.
      • Almeida L.L.
      • Ruffino-Netto A.
      • Kritski A.L.
      • Mello F.C.
      • Werneck G.L.
      Classification and regression tree (CART) model to predict pulmonary tuberculosis in hospitalized patients.
      • Westreich D.
      • Lessler J.
      • Funk M.J.
      Propensity score estimation: neural networks, support vector machines, decision trees (CART), and meta-classifiers as alternatives to logistic regression.
      • Takahashi O.
      • Cook E.F.
      • Nakamura T.
      • Saito J.
      • Ikawa F.
      • Fukui T.
      Risk stratification for in-hospital mortality in spontaneous intracerebral haemorrhage: a classification and regression tree analysis.
      The CART methodology with the Gini index rule was used for the following 2 models: model 1, a prediction model using a prerehabilitation data set; and model 2, a prediction model using a data set of prerehabilitation and 2 weeks after surgery (see table 1). The pruning rule in the CART included bootstrapped V-fold cross-validation using the minimum cross-validation error ± 1 standard error rule. A predicted rank was set by a positive likelihood ratio (PLR) of each subgroup. The accuracy of CART was evaluated using the area under the receiver operating characteristic curves (AUROCs) developed from each method. The maximum Youden Index (sensitivity + specificity − 1) was defined as the optimal cutoff point. The AUROC could distinguish between nonpredictive (AUROC<0.5), less predictive (0.5<AUROC<0.7), moderately predictive (0.7<AUROC<0.9), highly predictive (0.9<AUROC<1), and perfect prediction (AUROC=1).
      • Swets J.A.
      Measuring the accuracy of diagnostic systems.
      • Greiner M.
      • Pfeiffer D.
      • Smith R.D.
      Principles and practical application of the receiver-operating characteristic analysis for diagnostic tests.
      Cases with missing data were included for analysis. For patients lost to follow-up (dropout), data were used only to compare patient characteristics before rehabilitation and 2 weeks after surgery. All statistical analyses were performed using Stata version 13 software for Windows.a

      Study size

      Narita et al
      • Narita H.
      • Suetsuna F.
      • Fujii K.
      • Masutani M.
      • Shiozaki T.
      Treatment and prognosis of femoral neck fractures in the aged over 80-year-old patients.
      reported that 76.4% of elderly Japanese patients (65–79y of age) without cognitive impairment fully recovered mobility after surgery for hip fracture. Based on this finding, we assumed that the ratio of negative to positive patients (ie, those with ADL decline at 6mo after surgery) was 3 (75%) to 1 (25%). The alpha value was set at .05, and the power was set at 0.8. The hypothesized AUROC and the null hypothesis AUROC were set at 0.7 (meaning moderate power) and 0.5 (meaning no discriminating power), respectively. Consequently, 22 cases were required in the positive group and 66 in the negative group, for a total of 88 cases.

      Results

      At 6 months after surgery, 86 patients (82.6%) were known to be alive, 1 patient (1.0%) had died, and 17 (16.3%) were lost to follow-up. The reasons for discontinuation of follow-up were unclear because the investigators phoned the patients many times but were unable to make contact with the patients or their families.
      Data were missing for ADL before fracture or 2 weeks after surgery (1 case), CS-30 (1 case), and pain intensity (2 cases). We did not perform imputation of missing data, and these cases were excluded from the CART model 1 or 2. There were no significant differences in patient characteristics at 2 weeks after surgery between those who completed the 6-month follow-up and those who did not. Other main characteristics are summarized in table 2. ADL were not recovered at 6 months after surgery, as compared with baseline, in 32 (37.2%) of the 86 patients who completed the follow-up.
      Table 2Patient characteristics at prerehabilitation and 2 weeks after surgery
      TimeVariableCategoryAll (N=104)Completer (n=86)Dropout (n=18)P
      Prerehabilitation (baseline)Age (y)80.8±10.580.8±10.080.7±13.1.957
      Independent t test.
      SexMale25 (24)21 (24)4 (22).843
      Chi-square test.
      Female79 (76)65 (76)14 (78)
      Gait method before fractureIndependent gait67 (64)54 (63)13 (72).373
      Chi-square test.
      Crutch walking13 (13)11 (13)2 (11)
      Gait using a walker6 (6)6 (7)0 (0)
      Walking while holding on to something (such as a wall or table)18 (17)15 (17)3 (17)
      Dwelling placeHome91 (88)74 (86)17 (94).851
      Chi-square test.
      Institution13 (13)12 (14)1 (6)
      ADL before fractureTotal100 (40–100)100 (50–100)100 (40–100).686
      Mann-Whitney U test.
      Nursing care level before fractureNone61 (59)49 (57)12 (67).099
      Chi-square test.
      Needed support level 19 (9)7 (8)2 (11)
      Needed support level 28 (8)7 (8)1 (6)
      Care level 116 (15)14 (16)2 (11)
      Care level 25 (5)4 (5)1 (6)
      Care level 35 (5)5 (6)0 (0)
      Care level 40 (0)0 (0)0 (0)
      Care level 50 (0)0 (0)0 (0)
      Type of fractureFemoral neck67 (64)56 (65)11 (61).140
      Chi-square test.
      Trochanteric33 (32)27 (31)6 (33)
      Subtrochanteric4 (4)3 (3)1 (6)
      Surgical formFemoral head replacement53 (51)43 (50)10 (56).193
      Chi-square test.
      Gamma nail or alexa nail39 (38)32 (37)7 (39)
      Hansson pin4 (4)4 (5)0 (0)
      Others8 (8)7 (8)1 (6)
      Two weeks after surgeryADLFeeding10 (5–10)10 (0–10)10 (5–10).814
      Mann-Whitney U test.
      Chair transfer15 (0–15)15 (5–15)10 (0–15).266
      Mann-Whitney U test.
      Grooming5 (0–5)5 (5–5)5 (0–5).984
      Mann-Whitney U test.
      Toileting5 (0–10)5 (0–10)5 (0–10).851
      Mann-Whitney U test.
      Bathing0 (0–5)0 (0–5)0 (0–0).153
      Mann-Whitney U test.
      Ambulation10 (0–15)10 (0–15)10 (0–15).913
      Mann-Whitney U test.
      Stair climbing0 (0–10)0 (0–10)0 (0–10).095
      Mann-Whitney U test.
      Dressing5 (0–10)5 (0–10)5 (0–10).707
      Mann-Whitney U test.
      Bowel control10 (0–10)10 (0–10)10 (0–10).435
      Mann-Whitney U test.
      Bladder control10 (0–10)10 (0–10)10 (0–10).637
      Mann-Whitney U test.
      Total60 (10–100)60 (10–100)65 (10–90).584
      Mann-Whitney U test.
      CS-303.6±4.64.6±4.62.6±4.6.198
      Mann-Whitney U test.
      Pain3.5±2.73.3±2.64.6±2.9.106
      Mann-Whitney U test.
      NOTE. Values are mean ± SD, n (%), median (minimum–maximum), or as otherwise indicated.
      Independent t test.
      Chi-square test.
      Mann-Whitney U test.
      For model 1 (fig 1), the CART method identified age (<81y or ≥81y) as the best single discriminator for a decline in ADL at 6 months after surgery. Among those aged ≥81 years, the next best predictor was “type of fracture (femoral neck vs trochanteric or intertrochanteric).” Among those with “type of fracture: femoral neck,” the next predictor was age (<88y or ≥88y). Among those with “type of fracture: trochanteric or intertrochanteric,” the next predictor was care level, dichotomized at “needed support level 1.” The final tree model and 5 terminal nodes identified using CART analysis in model 1 are shown in figure 1 and table 3. The rank-1 group included 45.3% of patients (39/86), with a PLR of .248 and a positive predictive value (PPV) of 12.8%. The rank-2 group included 15.1% of patients (13/86), with a PLR of .506 and a PPV of 23.1%. The rank-3 group included 8.1% of patients (7/86), with a PLR of 1.266 and a PPV of 42.9%. The rank-4 group included 15.1% of patients (13/86), with a PLR of 3.797 and a PPV of 69.2%. The rank-5 group included 16.3% of patients (14/86), with a PLR of 10.125 and a PPV of 85.7%, which indicates excellent accuracy. Based on the AUROCs, the accuracy of the CART model 1 was .825 (95% confidence interval, .728–.923), with an optimal cutoff point of rank 3 or greater (sensitivity=75.0%, specificity=81.5%, PPV=70.6%, PLR=4.050).
      Fig 1
      Fig 1CART model 1 using prerehabilitation data. The CART method identified “age” as the best single discriminator for a decline in ADL after surgery. Other predictors were “type of fracture” and “nursing care level before fracture.”
      Table 3Accuracy of CART models and CPRs
      CPRsSensitivitySpecificityPPVNPVPLRAUROC (95% CI)Optimal Cutoff Point
      CART model 1: Prerehabilitation data (see fig 1).750.815.706.1544.05.825 (.728–.923)≥ Rank 3
       Rank 1“Age” <81.156.370.128.5740.248
       Rank 2“Age” ≥88 and “type of fracture” femoral neck.094.815.231.3970.506
       Rank 3“Age” ≥81 and “type of fracture” trochanteric or subtrochanteric and “nursing care level before fracture” none.094.926.429.3671.266
       Rank 4“Age” ≥81 and <88 and “type of fracture” femoral neck.281.926.692.3153.797
       Rank 5“Age” ≥81 and “type of fracture” trochanteric or subtrochanteric and “nursing care level before fracture” ≥ needed support level 1.375.963.857.27810.125
      CART model 2: Prerehabilitation data + data of 2nd week after surgery (see fig 2).656.870.750.1905.063.790 (.683–.897)≥ Rank 3
       Rank 1“ADL chair transfer 2nd week after surgery” =15.250.259.167.6320.338
       Rank 2“ADL chair transfer 2nd week after surgery” <15 and “ADL ambulation 2nd week after surgery” ≥5 and “age” <83.094.870.300.3820.723
       Rank 3“ADL chair transfer 2nd week after surgery” <15 and “ADL ambulation 2nd week after surgery” ≥5 and “age” ≥83.250.926.667.3243.375
       Rank 4“ADL chair transfer 2nd week after surgery” <15 and “ADL ambulation 2nd week after surgery” <5.406.944.813.2717.313
      NOTE. Optimal cutoff point was determined using Youden Index.
      Abbreviations: CI, confidence interval; NPV, negative predictive value.
      For model 2 (fig 2), the CART method identified “postsurgery 2 weeks ADL chair transfer” (≤15) as the best single discriminator for the decline in ADL at 6 months. For those with a “postsurgery 2 weeks ADL chair transfer” of <15, the next best predictor was “postsurgery 2 weeks ADL ambulation (≥5 or <5).” Among those with a “postsurgery 2 weeks ADL ambulation” ≥5, the next predictor was age, dichotomized at 83 years. The final tree model for model 2 with the 4 terminal nodes identified by the CART analysis is shown in figure 2 and table 3. The rank-1 group included 55.8% of patients (48/86), with a PLR of .338 and a PPV of 16.7%. The rank-2 group included 11.6% of patients (10/86), with a PLR of .723 and a PPV of 30.0%. The rank-3 group included 14.0% of patients (12/86), with a PLR of 3.375 and a PPV of 66.7%. The rank-4 group included 18.6% of patients (16/86), with a PLR of 7.313 and a PPV of 81.3%. Based on the AUROCs, the accuracy of the CART model 2 was .790 (95% confidence interval, .683–.897), with an optimal cutoff point of rank 3 or greater (sensitivity=65.6%, specificity=87.0%, PPV=75.0%, PLR=5.063).
      Fig 2
      Fig 2CART model 2 using prerehabilitation data plus data from 2nd week after surgery. The CART method identified “ADL chair transfer 2nd week after surgery” as the best single discriminator for a decline in ADL after surgery. Other predictors were “ADL ambulation after 2nd week after surgery” and “age.”

      Discussion

      We developed and assessed a CPR to predict declines in ADL at 6 months after surgery for hip fracture repair. Therefore, these CPRs had moderate levels of accuracy.
      Age,
      • Penrod J.D.
      • Litke A.
      • Hawkes W.G.
      • et al.
      Heterogeneity in hip fracture patients: age, functional status, and comorbidity.
      • Haentjens P.
      • Autier P.
      • Barette M.
      • Boonen S.
      Belgian Hip Fracture Study Group
      Predictors of functional outcome following intracapsular hip fracture in elderly women. A one-year prospective cohort study.
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      • Eastwood E.A.
      • Magaziner J.
      • Wang J.
      • et al.
      Patients with hip fracture: subgroups and their outcomes.
      type of fracture,
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      and ADL
      • Penrod J.D.
      • Litke A.
      • Hawkes W.G.
      • et al.
      Heterogeneity in hip fracture patients: age, functional status, and comorbidity.
      • Eastwood E.A.
      • Magaziner J.
      • Wang J.
      • et al.
      Patients with hip fracture: subgroups and their outcomes.
      • Kagaya H.
      • Takahashi H.
      • Sugawara K.
      • Dobashi M.
      • Kiyokawa N.
      • Ebina H.
      Predicting outcomes after hip fracture repair.
      • Alegre-Lopez J.
      • Cordero-Guevara J.
      • Alonso-Valdivielso J.L.
      • Fernandez-Melon J.
      Factors associated with mortality and functional disability after hip fracture: an inception cohort study.
      before rehabilitation were significantly predictive of a decline in ADL at over 6 months after surgery for hip fracture repair. Although some studies have suggested the power of various factors to predict a decline in ADL, the accuracy of these predictions was not examined. We showed that age, type of fracture, and prerehabilitation ADL were predictors for a decline in ADL at 6 months after surgery, in accordance with the findings of previous studies.
      • Penrod J.D.
      • Litke A.
      • Hawkes W.G.
      • et al.
      Heterogeneity in hip fracture patients: age, functional status, and comorbidity.
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      • Eastwood E.A.
      • Magaziner J.
      • Wang J.
      • et al.
      Patients with hip fracture: subgroups and their outcomes.
      • Kagaya H.
      • Takahashi H.
      • Sugawara K.
      • Dobashi M.
      • Kiyokawa N.
      • Ebina H.
      Predicting outcomes after hip fracture repair.
      • Alegre-Lopez J.
      • Cordero-Guevara J.
      • Alonso-Valdivielso J.L.
      • Fernandez-Melon J.
      Factors associated with mortality and functional disability after hip fracture: an inception cohort study.
      Furthermore, we developed a CPR with a moderate precision for predicting a decline in ADL 6 months after surgery. CART analysis is useful to identify predictive factors while considering mutual relationships among factors with high contribution ratios. In this study, the best predictive factor identified by model 1 was age, followed by the type of fracture and nursing care level. Thus, elderly patients who have trochanteric or intertrochanteric fractures and who needed support for their ADL before the fracture are considered to be at a higher risk for a decline in ADL at 6 months after surgery. This reliable CPR should prove useful to rehabilitation therapists to screen for patients who are at greater risk for a decline in ADL after surgery.
      Most previous studies did not identify factors during hospitalization to predict ADL decline at over 6 months after surgery. Therefore, rehabilitation therapists have limited information on what factors should be targeted to prevent a decline in ADL after discharge. The model 2 results showed that the level of independent chair transfer and ambulation, as assessed using the Barthel Index at 2 weeks after surgery, and age are significant predictive factors for the recovery of ADL. In addition, the precision of model 2, as well as model 1, was supported by a high PLR and AUROC. Although health care providers cannot control age, type of fracture, or nursing care level before rehabilitation, the level of independence in performing a chair transfer and in ambulating at 2 weeks after surgery may be improved by early intervention through rehabilitation programs. Along with the development of a CPR, we identified the cutoff values of predictors using CART analysis. We believe that in early rehabilitation, these values should be targeted at 2 weeks after surgery.
      Internationally, CART has become increasingly prevalent because of the sentinel work by Breiman et al.
      • Breiman L.
      • Friedman J.
      • Olshen R.
      • Stone C.
      Classification and regression trees.
      CART is an exploratory research method to uncover relationships and produce clearly illustrated associations between variables not amenable to traditional linear regression analysis.
      • Crichton N.J.
      • Hinde J.P.
      • Marchini J.
      Models for diagnosing chest pain: is CART helpful?.
      This method has a long history in market research and has more recently become increasingly used in medicine to stratify risks
      • Karaolis M.A.
      • Moutiris J.A.
      • Hadjipanayi D.
      • Pattichis C.S.
      Assessment of the risk factors of coronary heart events based on data mining with decision trees.
      and determine prognoses.
      • Lamborn K.R.
      • Chang S.M.
      • Prados M.D.
      Prognostic factors for survival of patients with glioblastoma: recursive partitioning analysis.
      A literature review of studies that used CART revealed relatively few studies, with only 1 study
      • Arai T.
      • Kaneko S.
      • Fujita H.
      [Decision trees on gait independence in patients with femoral neck fracture].
       written in Japanese. Compared with the statistical methods used in previous studies, our proposed novel approach is expected to be used in future studies for the development of CPRs.
      Generally, the elderly are at a greater risk of femoral neck fractures and mild walking disabilities, and some may have cognitive impairments. If possible, data should be collected from various patients to generalize CPRs developed by this study. However, cognitive impairment is another problem associated with locomotive impairment. Accordingly, we targeted elderly patients with femoral neck fractures, independent locomotion, and no serious cognitive impairment before fracture. Hence, our study findings can be generalized to patients who meet our inclusion criteria with neither of the exclusion criteria. In addition, our investigation period was 6 months. It is unknown whether these findings can be applied to predict a decline in ADL over 6 months after surgery. We continue to follow-up the study participants for further verification of this model in future studies.

      Study limitations

      There were several limitations to this study that should be addressed. First, there was a potential selection bias because patients with serious complications immediately after surgery may have hesitated to participate in this study, compared with those who fared better. Patients with serious complications immediately after surgery are most likely to recover more slowly, with ADL levels at 6 months after surgery lower than those before fracture. In general, this tendency is particularly prevalent among elderly patients. The results of this study identified age as a risk factor for a decline in ADL at 6 months after surgery in both CPR models 1 and 2. If this bias could be reduced, the number of patients not recovering ADL would increase, and both PPV and PLR would positively change. Second, the strength of confounding factors on our results was not adjusted. We believe this limitation is not important because CART analysis considers mutual relationships among factors included in the decision tree model. However, this study did not address psychological factors such as mental status
      • Parker M.J.
      • Palmer C.R.
      Prediction of rehabilitation after hip fracture.
      • Alegre-Lopez J.
      • Cordero-Guevara J.
      • Alonso-Valdivielso J.L.
      • Fernandez-Melon J.
      Factors associated with mortality and functional disability after hip fracture: an inception cohort study.
      and fear of falling,
      • Oude Voshaar R.C.
      • Banerjee S.
      • Horan M.
      • et al.
      Fear of falling more important than pain and depression for functional recovery after surgery for hip fracture in older people.
      which have already been confirmed as risk factors for a decline in ADL. Although the predictors used were selected based on the results of previous studies, these studies may not have included all possible predictors. For example, it appears as though psychosocial variables have not been adequately explored. Alternative CPRs that include these factors may be more accurate than CPRs proposed in the present study. Another possible limitation was the scoring method of the Barthel Index. The scoring was based on objective assessments during hospitalization, but on self-reports at 6 months after surgery. Thus, there may have been a reporting bias in our data. To minimize this bias, in future studies, the method for scoring the Barthel Index should be integrated with objective assessment by the same trained assessor.

      Conclusions

      CPRs with moderate accuracy were developed to predict declines in ADL at 6 months after surgery in patients with hip fractures. These CPRs are reliable and is worthwhile for the screening of patients at a greater risk of ADL decline after surgery. For preventing the ADL decline, elderly patients who have trochanteric or intertrochanteric fractures and who needed support for their ADL before the fracture should receive more intensive rehabilitation, more timely rehabilitation, or both, to recover their ability to chair transfer and ambulate by 2 weeks after surgery.

      Supplier

      • a.
        Stata version 13 software for Windows; StataCorp LP.

      Acknowledgments

      We thank Nobuhiro Kito, PhD (Department of Rehabilitation, Faculty of Rehabilitation, Hiroshima International University), for his advice regarding the method of data collection, and Akihiro Toyota, PhD, Mitsuo Tagami, MA, Kenji Nishiyama, Katsumi Toba, Ikue Uemura, Maki Hara, Yasutaka Shingai, Takuya Hayashizaki, Naoto Wada, Rika Ishii, and Susumu Nagao (Chūgokurōsai Hospital and Saiseikaikure Hospital, Kure, Japan) for their support in data collection.

      Appendix

      Appendix 1Independence Degree of Daily Living for the Disabled Elderly
      Rank JSome disabilities, but daily living is mostly independent; capable of going outdoors unassisted.
      1. Goes outdoors with means of transportation.
      2. Goes out near home.
      Rank AIndoor living predominantly independent, but unable to go out without assistance.
      1. Goes out with assistance, spending most time during the daytime out of bed.
      2. Does not go out frequently, repeating cycles of lying down on and getting up from a bed during the daytime.
      Rank BSome assistance needed for indoor living; also lies in bed for much of the daytime, although sitting position is possible.
      1. Uses a wheelchair without assistance, takes meals, and excretes/urinates off the bed.
      2. Uses a wheelchair with assistance.
      Rank CBedridden all day; requires assistance with excretion/urination, meals, and dressing/undressing.
      1. Capable of changing posture in bed.
      2. Unable to change posture in bed without assistance.
      Appendix 2Independence Degree of Daily Living for the Elderly With Dementia
      Rank IHas some type of dementia, but almost independent in terms of daily living at home and in society.
      Rank IISome daily life-disturbing symptoms, behaviors, and problems in communication seen, but can lead daily life independently if watched by someone.
       IIaCondition II, mentioned above, seen outside home.
       IIbCondition II, mentioned above, seen at home.
      Rank IIIDaily life-disturbing symptoms, behaviors, and problems in communication that require assistance.
       IIIaCondition III, mentioned above, seen primarily during the daytime.
       IIIbCondition III, mentioned above, seen primarily at night.
      Rank IVDaily life-disturbing symptoms, behaviors, and problems in communication frequently require assistance.
      Rank MMarked psychiatric symptoms/related symptoms or serious physical disorders require expert management.

      References

        • Johnell O.
        • Kanis J.
        Epidemiology of osteoporotic fractures.
        Osteoporos Int. 2005; 16: S3-S7
        • Sambrook P.
        • Cooper C.
        Osteoporosis.
        Lancet. 2006; 367: 2010-2018
        • Morrison R.S.
        • Chassin M.R.
        • Siu A.L.
        The medical consultant's role in caring for patients with hip fracture.
        Ann Intern Med. 1998; 128: 1010-1020
        • Johnell O.
        The socioeconomic burden of fractures: today and in the 21st century.
        Am J Med. 1997; 103 (discussion 25S-26S): 20S-25S
        • Penrod J.D.
        • Litke A.
        • Hawkes W.G.
        • et al.
        Heterogeneity in hip fracture patients: age, functional status, and comorbidity.
        J Am Geriatr Soc. 2007; 55: 407-413
        • Haentjens P.
        • Autier P.
        • Barette M.
        • Boonen S.
        • Belgian Hip Fracture Study Group
        Predictors of functional outcome following intracapsular hip fracture in elderly women. A one-year prospective cohort study.
        Injury. 2005; 36: 842-850
        • Parker M.J.
        • Palmer C.R.
        Prediction of rehabilitation after hip fracture.
        Age Ageing. 1995; 24: 96-98
        • Eastwood E.A.
        • Magaziner J.
        • Wang J.
        • et al.
        Patients with hip fracture: subgroups and their outcomes.
        J Am Geriatr Soc. 2002; 50: 1240-1249
        • Kagaya H.
        • Takahashi H.
        • Sugawara K.
        • Dobashi M.
        • Kiyokawa N.
        • Ebina H.
        Predicting outcomes after hip fracture repair.
        Am J Phys Med Rehabil. 2005; 84: 46-51
        • Alegre-Lopez J.
        • Cordero-Guevara J.
        • Alonso-Valdivielso J.L.
        • Fernandez-Melon J.
        Factors associated with mortality and functional disability after hip fracture: an inception cohort study.
        Osteoporos Int. 2005; 16: 729-736
        • Ingemarsson A.H.
        • Frandin K.
        • Mellstrom D.
        • Moller M.
        Walking ability and activity level after hip fracture in the elderly—a follow-up.
        J Rehabil Med. 2003; 35: 76-83
        • Shyu Y.I.
        • Chen M.C.
        • Liang J.
        • Wu C.C.
        • Su J.Y.
        Predictors of functional recovery for hip fractured elders during 12 months following hospital discharge: a prospective study on a Taiwanese sample.
        Osteoporos Int. 2004; 15: 475-482
        • Kim S.M.
        • Moon Y.W.
        • Lim S.J.
        • et al.
        Prediction of survival, second fracture, and functional recovery following the first hip fracture surgery in elderly patients.
        Bone. 2012; 50: 1343-1350
        • Shyu Y.I.
        • Chen M.C.
        • Wu C.C.
        • Cheng H.S.
        Family caregivers' needs predict functional recovery of older care recipients after hip fracture.
        J Adv Nurs. 2010; 66: 2450-2459
        • Shyu Y.I.
        • Tang W.R.
        • Tsai W.C.
        • Liang J.
        • Chen M.C.
        Emotional support levels can predict physical functioning and health related quality of life among elderly Taiwanese with hip fractures.
        Osteoporos Int. 2006; 17: 501-506
        • Oude Voshaar R.C.
        • Banerjee S.
        • Horan M.
        • et al.
        Fear of falling more important than pain and depression for functional recovery after surgery for hip fracture in older people.
        Psychol Med. 2006; 36: 1635-1645
        • McGinn T.G.
        • Guyatt G.H.
        • Wyer P.C.
        • Naylor C.D.
        • Stiell I.G.
        • Richardson W.S.
        Users' guides to the medical literature: XXII: how to use articles about clinical decision rules. Evidence-Based Medicine Working Group.
        JAMA. 2000; 284: 79-84
        • Harris W.H.
        Traumatic arthritis of the hip after dislocation and acetabular fractures: treatment by mold arthroplasty. An end-result study using a new method of result evaluation.
        J Bone Joint Surg Am. 1969; 51: 737-755
        • Mahoney F.I.
        • Barthel D.W.
        Functional evaluation: the Barthel Index.
        Md State Med J. 1965; 14: 61-65
        • Jones C.J.
        • Rikli R.E.
        • Beam W.C.
        A 30-s chair-stand test as a measure of lower body strength in community-residing older adults.
        Res Q Exerc Sport. 1999; 70: 113-119
        • Della Pietra G.L.
        • Savio K.
        • Oddone E.
        • Reggiani M.
        • Monaco F.
        • Leone M.A.
        Validity and reliability of the Barthel Index administered by telephone.
        Stroke. 2011; 42: 2077-2079
        • Breiman L.
        • Friedman J.
        • Olshen R.
        • Stone C.
        Classification and regression tree analysis.
        Boston Univ Tech Rep. 2014; 1: 1-16
        • Aguiar F.S.
        • Almeida L.L.
        • Ruffino-Netto A.
        • Kritski A.L.
        • Mello F.C.
        • Werneck G.L.
        Classification and regression tree (CART) model to predict pulmonary tuberculosis in hospitalized patients.
        BMC Pulm Med. 2012; 12: 40
        • Westreich D.
        • Lessler J.
        • Funk M.J.
        Propensity score estimation: neural networks, support vector machines, decision trees (CART), and meta-classifiers as alternatives to logistic regression.
        J Clin Epidemiol. 2010; 63: 826-833
        • Takahashi O.
        • Cook E.F.
        • Nakamura T.
        • Saito J.
        • Ikawa F.
        • Fukui T.
        Risk stratification for in-hospital mortality in spontaneous intracerebral haemorrhage: a classification and regression tree analysis.
        QJM. 2006; 99: 743-750
        • Swets J.A.
        Measuring the accuracy of diagnostic systems.
        Science. 1988; 240: 1285-1293
        • Greiner M.
        • Pfeiffer D.
        • Smith R.D.
        Principles and practical application of the receiver-operating characteristic analysis for diagnostic tests.
        Prev Vet Med. 2000; 45: 23-41
        • Narita H.
        • Suetsuna F.
        • Fujii K.
        • Masutani M.
        • Shiozaki T.
        Treatment and prognosis of femoral neck fractures in the aged over 80-year-old patients.
        J Eastern Japan Assoc Orthop Traumatol. 2003; 15: 194-197
        • Breiman L.
        • Friedman J.
        • Olshen R.
        • Stone C.
        Classification and regression trees.
        Wadsworth International Group, Belmont1984
        • Crichton N.J.
        • Hinde J.P.
        • Marchini J.
        Models for diagnosing chest pain: is CART helpful?.
        Stat Med. 1997; 16: 717-727
        • Karaolis M.A.
        • Moutiris J.A.
        • Hadjipanayi D.
        • Pattichis C.S.
        Assessment of the risk factors of coronary heart events based on data mining with decision trees.
        IEEE Trans Inf Technol Biomed. 2010; 14: 559-566
        • Lamborn K.R.
        • Chang S.M.
        • Prados M.D.
        Prognostic factors for survival of patients with glioblastoma: recursive partitioning analysis.
        Neuro Oncol. 2004; 6: 227-235
        • Arai T.
        • Kaneko S.
        • Fujita H.
        [Decision trees on gait independence in patients with femoral neck fracture].
        Nihon Ronen Igakkai Zasshi. 2011; 48 ([Japanese]): 539-544