Factors associated with change in pain and disability over time: a community-based prospective observational study of hip and knee osteoarthritis

Abstract

Background

Hip and knee osteoarthritis are frequent causes of primary care consultations. They are considered slowly progressive disorders, often resulting in severe pain or disability and the need for joint replacements. There have been few longitudinal studies of progression to inform individual prognoses in primary care.

Aim

To describe the degree of progression and investigate predictors of change in hip or knee pain and disability.

Design of study

Prospective community-based cohort.

Setting

An age–sex stratified survey of 27 000 people registered with 40 general practices in Avon and Somerset yielded 2437 reporting hip and/or knee symptoms at baseline (1992–1994). A 25% random sample of 587 individuals was followed up between 1998 and 1999.

Method

Pain or disability was measured at baseline and follow-up using the New Zealand score. For the worst joint according to the New Zealand score at baseline, hip and knee problems were analysed separately. Regression models ascertained characteristics of non-responders and factors associated with change in scores. Seven sociodemographic, seven comorbidity, and two healthcare utilisation variables were considered.

Results

Generally pain and disability worsened over the 7 years, but 35% and 29% of those initially reporting hip and knee pain respectively had improved. Reporting ‘other health problems’ was associated with greater deterioration for both hip and knee disease, as was cardiovascular morbidity for hip disease and lower social class, being retired, hypertension, and higher body mass index for knee disease. Deteriorations in scores were strongly associated with individuals consulting their GP about joint problems.

Conclusion

Osteoarthritis does not invariably deteriorate, but when it does social as well as biological factors appear to be important. These findings may aid outcome prediction. Future research on osteoarthritis should be conducted within a biopsychosocial rather than a purely biological paradigm.

INTRODUCTION

Osteoarthritis (OA) of the hip and knee are common conditions in the community and a frequent cause of primary care consultations where people present with regional pain or disability.^1-3 It is generally believed that OA progresses gradually, and that it is often accompanied by severe pain and disability, eventually requiring joint replacement. The few previous longitudinal studies of symptom progression suggest a much more heterogeneous outcome, with some patients improving and only a minority needing surgery.⁴ Knowledge is lacking on which to base individual prognoses.

Here, a community-based cohort of people who reported hip or knee problems as part of a detailed baseline assessment were followed up 7 years later to establish levels of deterioration of pain and function. The purpose was to investigate factors that might predict change and thus help GPs provide a realistic prognosis for the large number of their patients with this disease. The choice of predictor variables was focused on simple factors thought likely to affect the outcome of OA based on the current literature. Age, sex, body mass index (BMI) and comorbidities have been highlighted as likely factors⁵ and cardiovascular disease has been identified as a possible comorbidity modulator of pain and progression in OA.^6,7 Psychosocial factors are also known to affect pain and disability^8,9 and to influence outcome for those with self-reported generic chronic illness.¹⁰

How this fits in

Hip and knee osteoarthritis are frequent causes of primary care consultations. Very little is known either about the degree of progression or regression, or the factors that may be helpful in making individual prognoses in primary care. Reported cardiovascular problems and general morbidity are associated with greater deterioration for both hip and knee disease, as is social disadvantage and higher body mass index for knee disease. The evidence presented here may be useful in anticipating potential patient outcomes. Any further research in this field should acknowledge the influence of psychosocial as well as biological factors.

The assessment tool used was the New Zealand (NZ) score,¹¹ developed to aid prioritisation for surgery and advocated for routine use within primary care as a means to prioritise referrals for joint replacement.¹² An ancillary aim of our research was to examine the sensitivity to change of the NZ score, which has not previously been used in longitudinal studies.

METHOD

Participants

The Somerset and Avon Survey of Health (SASH) is a community-based, age–sex stratified survey of 27 000 people registered with 40 general practices between 1992 and 1994.^13,14 Briefly, a postal screening questionnaire was sent to an age–sex stratified random sample of individuals from 40 practices. This multistage sampling process introduces clustering effects by general practice, the selection of which took into account practice size.¹⁵ Within the screening questionnaire were items relating to joint pain; for example, ‘During the past 12 months, have you had pain in or around either of your [hips/knees] on most days for 1 month or longer?’ Individuals reporting such symptoms were invited for further assessment at a clinic or during a home visit, which included a full clinical examination of the hip, knee, and lower back.^13,14

A total of 2437 screening questionnaire responders reported hip and/or knee symptoms, and these data were used to calculate NZ scores (Table 1).¹¹ Preserving the original sampling design (clustered by practice; stratified by sex and age) and with a further stratification into three clinical groups (those reporting hip problems, knee problems, or both), a 25% random sample of 587 individuals was followed up between 1998 and 1999.

Table 1.

Criteria for the New Zealand Score.^a

Clinical features	Maximum score allotted to feature
1. Pain	40
2. Functional activity
Walking	10
Activities of daily living (washing, dressing, etc)	10
3. Movement and deformity (based on clinical examination and x-ray)	20b
4. Other factors
Multiple joint disease	10
Ability to work, meet caring commitments, live independently	10

^aThis table is adapted from Hadorn DC, Holmes AC. The New Zealand priority criteria project. Part 1: Overview. BMJ 1997; 314: 131–134 (Table 2). Amended and reproduced with kind permission from the BMJPublishing Group.

^bAs this feature was only available at baseline, for consistency it was not used here; scores were therefore out of 80 both at baseline and follow-up.

Sampled individuals were invited to participate by letter, and those agreeing were asked identical questions to those in the original survey by telephone. Non-responders were reminded with a second letter. Factors associated with non-response were investigated using logistic regression taking into account stratification and clustering (Stata Statistical Software: Release 8.0).

Scoring and selection for the analyses

The NZ score¹¹ allocates points out of 100 (Table 1). In this study we compared the change in scores between baseline and follow-up, each measured out of a maximum (worst) score of 80 (excluding the clinical examination as it was not conducted at follow-up). For both hips and knees, thresholds of 43 and 55 out of 100 are considered to reflect moderate and severe disease respectively,^13,14 corresponding to 35 and 44 points out of 80. Each responder was included in the relevant (hip or knee) analyses for their worst joint (left or right) at baseline. Where scores for both joints were the same or zero, the worst joint at follow-up was taken for analysis of change between baseline and follow-up.

Factors related to (change in) NZ scores

Descriptive statistics were obtained for the NZ score of the worst joint at baseline and follow-up, together with the change in scores between the two time points. With the worst hip and the worst knee considered in separate analyses,¹⁴ potential predictor variables were included in ordinary regression analyses of the NZ score as a continuous measure at follow-up, adjusted for baseline. Univariable and multivariable regression models ascertained the independent relationships with the outcome (follow-up NZ score) for the 14 variables, adjusting in all models for the relevant baseline score and the sampling design, that is, stratification by sex and age, and clustering by practice.

The 14 explanatory variables were considered in two groups of similar factors: seven sociodemographic characteristics (sex, age group, body mass index [BMI in kg/m²], employment status [self and partner], social class, and marital status) and seven binary (yes/no) comorbidity variables (cardiovascular, respiratory and eye disease, hypertension, depression, cancer, other health problems). Two binary healthcare utilisation variables (consulting a GP or hospital doctor about joint problems) were considered separately as they would likely be affected by the outcome (disease severity) rather than influence it. Independent relationships between the outcome and the other 14 factors were, therefore, ascertained both before and after considering healthcare utilisation. Only linear relationships between BMI and NZ score are presented as this assumption was reasonable after considering quadratic terms and BMI quartiles.

Factors associated with follow-up NZ scores at the 10% significance level in univariable models were entered into multivariable regression analyses to identify variables independently associated with change (relaxing the initial threshold to 20% ultimately made no difference). Using a stricter 5% threshold for statistical significance, multivariable models first accounted for inter-relationships within each of the two groups of factors separately and then combined. The final models derived from this procedure were then reconsidered after adjusting for healthcare utilisation. For all final models the assumption of normally distributed residuals was confirmed by standard graphical methods; the R²-statistic assessed goodness-of-fit.

To allow for effects of surgery on the relevant joint between baseline and follow-up, the final models were subjected to two sensitivity analyses. First, surgery was adjusted for by including the appropriate (binary) variable in each model. Second, the NZ score at follow-up (likely to be lower than if intervention had not occurred) was replaced by 44 for those who received surgery, assuming a severe level of disease prior to surgery.

RESULTS

Of the 587 individuals sampled, 390 (66%) responded to the questionnaire. Of the 197 non-responders, 12 had died, six could not be traced, 128 refused to be included and a further 51 did not respond following one reminder. Of the 390 responders, 93 (24%) had hip problems only, 176 (45%) knee problems only and 121 (31%) both. Totals of 214 and 297 individuals were available for the analyses of the hip and knee NZ scores respectively, including 24 with hip and 25 with knee operations.

Characteristics of non-responders

The only variables independently associated with response rate were: age, social class and eye disease. The youngest (35–49 years) and oldest (≥70 years) age groups were least likely to respond (60% compared with nearly 75% for the middle age group; P = 0.0016). Those in social classes IV and V (about 50%) were less likely to respond than those in the other social classes (65% versus 83%; P = 0.0038), and those reporting any form of eye disease were less likely to respond than those without such problems (56% versus 69%; P = 0.036). Among the responders there was very little missing data, with the exception of BMI, known on about 85% of the individuals.

Change in NZ scores for those with hip arthropathies

The mean (95% confidence interval [CI]) NZ score for the worst hip was 22.5 (19.8 to 25.2) at baseline and 27.7 (25.1 to 30.2) at follow-up, with a mean deterioration of 5.2 (2.5 to 7.9). Individual differences ranged from an improvement of 42 to a deterioration of 48 points; 75 individuals, or 35% (95% CI = 28 to 42), improved between baseline and follow-up (32% excluding those operated on), with 20 (9% of all 214) showing no change and another 11 (5%) deteriorating by less than five points. At baseline, 29 (14%) would be considered to have ‘moderate’ and 31 (14%) ‘severe’ disease by the above thresholds; at follow-up these figures stood at 23 (11%) and 51 (24%) respectively.

Women deteriorated more than men (Table 2), but this could be by chance (adjusted difference = 4.4, 95% CI = −0.4 to 9.1; P = 0.070). Similarly for the changes in score by age group, where deterioration was greatest for the youngest and oldest age groups (Table 2; P = 0.28). Univariable regressions for the five other sociodemographic variables yielded: BMI (P = 0.14); employment status (P = 0.044); social class (P = 0.22); partner in paid employment – yes/no (P = 0.75); marital status – partner/no partner/widowed (P = 0.12). Hence, only employment status was included in the multivariable models. Of the seven comorbidity variables, four were not considered further: respiratory disease (P = 0.55), eye disease (P = 0.86), hypertension (P = 0.44), and cancer (P = 0.41). Univariable regressions for the other three yielded: cardiovascular morbidity (P = 0.024); depression (P = 0.038), and ‘other health problems’ (P = 0.017). Multivariable modelling among these three indicated that depression (adjusted P = 0.083) was partially confounded by the other two variables.

Table 2.

New Zealand hip scores by sex and age group, adjusting for stratification and clustering^a (n = 214).

Group (n)	Baseline		Follow-up		Difference [follow-up minus baseline]
	Mean score	95% CI	Mean score	95% CI	Mean score	95% CI
Females (138)	23.0	19.6 to 26.4	29.5	26.2 to 32.8	6.5	3.2 to 9.7
Males (76)	21.5	17.3 to 25.7	24.3	20.5 to 28.2	2.8	−2.0 to 7.6
Age 35–49 years (32)	17.2	9.9 to 24.3	23.8	17.3 to 30.2	6.6	0.5 to 12.7
Age 50–69 years (130)	23.2	19.6 to 26.7	27.0	23.6 to 30.5	3.9	0.5 to 7.3
Age ≥70 years (52)	24.0	19.1 to 28.3	31.7	27.0 to 36.4	7.7	1.7 to 13.6

^aUsing the ‘survey’ suite of programs in Stata.

Combining the three remaining sociodemographic and comorbidity variables led to the association for employment status being reduced both in magnitude and statistical significance (P = 0.20), and hence to the final model in Table 3. While sex and age group (categorised as in Table 2) were included in this model, for simplicity the results for these stratification variables are omitted from Table 3 (P = 0.11 and P = 0.82 respectively in this model). Adjusted for baseline, those who reported cardiovascular morbidity and ‘other health problems’ had higher (worse) NZ scores at follow-up compared with those not reporting such morbidity.

Table 3.

Final multiple regression model for New Zealand score in hips (n = 194, R² = 0.23^a).

Variableb	Category/scale	Number in category	Adjusted difference in means	95% CI	P-value
Cardiovascular morbidity					0.012
	No (reference)	97	0
	Yes	97	7.1	1.6 to 12.6
Other health problemsc					0.023
	No (reference)	147	0
	Yes	49	5.1	0.7 to 9.6

^a0.18 of this is attributable to baseline, sex and age.

^bAdjusted as before for the sampling design (stratification by sex and age; clustering by general practice).

^cOther than those specified in the list given in the methods section.

Ninety-seven (45%) responders stated that they had consulted a GP about their hip problems and 54 (25%) a hospital doctor between baseline and follow-up. Adjusting for sex and age group, these responders had deteriorated more than their counterparts (P <0.001 and P = 0.026 for consulting a GP and a hospital doctor, respectively). However, due to the role of GPs in referring patients to secondary care, the latter association was entirely removed (P = 0.44) after accounting for having seen a GP. The association between deterioration in NZ score and ‘other health problems’ was attenuated (P = 0.13) after adjusting for the binary GP consultation variable, whereas that for cardiovascular morbidity was little changed. Even in adjusting for the latter as well as sex and age group, those who had seen a GP had NZ scores that had deteriorated by, on average, 15.0 points more than those who had not (95% CI = 10.4 to 19.7; P <0.001; n = 214, R² = 0.34 for this four-variable model).

Change in NZ scores for those with knee arthropathies

The mean (95% CI) NZ score for the worst knee was 21.8 (19.9 to 23.6) at baseline and 28.8 (26.7 to 30.9) at follow-up, with a mean deterioration of 7.0 (5.0 to 9.1). Individual differences ranged from an improvement of 42 to a deterioration of 54 points; 87 individuals, or 29% (95% CI = 24 to 35), improved between baseline and follow-up (27% excluding those operated on), with 30 (10% of the total) showing no change and another 21 (7%) deteriorating by less than five points. At baseline, 33 (11%) and 36 (12%) had moderate and severe disease respectively, 40 (14%) and 78 (26%) at follow-up.

Women again deteriorated more than men (Table 4; adjusted difference = 3.6, 95% CI = −0.5 to 7.6; P = 0.087), and for knees the NZ score worsened to a greater degree with increasing age (Table 4; P = 0.0059). Although marital status was not associated with knee NZ score controlling for baseline and study design (P = 0.48), there were univariable associations for the other four sociodemographic factors: BMI (P = 0.041); employment status (P = 0.0039); social class (P = 0.038); and partner in paid employment (P = 0.030). Controlling for each other left just the first three of these for further modelling (P = 0.42 for partner's employment status adjusting for the others).

Table 4.

New Zealand knee scores by sex and age group, adjusting for stratification and clustering^a(n = 297).

Group (n)	Baseline		Follow-up		Difference [follow-up minus baseline]
	Mean score	95% CI	Mean score	95% CI	Mean score	95% CI
Females (174)	23.3	20.9 to 25.7	31.2	28.6 to 33.8	7.9	5.6 to 10.2
Males (123)	19.6	16.8 to 22.3	25.4	22.0 to 28.9	5.9	2.1 to 9.6
Age 35–49 years (38)	17.7	12.7 to 22.8	22.4	17.5 to 27.3	4.7	–0.2 to 9.6
Age 50–69 years (160)	20.9	18.6 to 23.3	27.1	24.1 to 30.1	6.1	3.0 to 9.3
Age ≥70 years (99)	24.6	21.3 to 28.0	34.1	30.1 to 37.6	9.5	6.4 to 12.5

^aUsing the ‘survey’ suite of programmes in Stata.

Univariable analyses suggested that hypertension (P = 0.016), depression (P = 0.051), and ‘other health problems’ (P = 0.037) were associated with change in NZ score. There was very weak evidence of an association for eye disease (P = 0.15) and no evidence for the other three morbidity variables (P = 0.32, 0.44 and 0.56 for cardiovascular, respiratory disease, and cancer respectively). Multivariable modelling within this group indicated that, as with the hip analyses, depression was attenuated (adjusted P = 0.13) by the remaining morbidity variables – in this case, hypertension and ‘other health problems’.

Combining the sociodemographic and morbidity variables gave the final model as shown in Table 5, again omitting sex and age group from the presentation (P = 0.013 and P = 0.75, respectively). Through this modelling, there was some attenuation of the associations for employment status and BMI by adjusting for the two morbidity variables, but those for the other three variables in Table 5 were unaltered. From Table 5, compared with the other groups, there were greater deteriorations among the retired (and possibly those sick/disabled although numbers were small), and a general trend towards worse knee NZ scores as social class decreased. Those with hypertension and ‘other health problems’ were considerably worse off than those without; again, these differences were virtually unaffected by allowing for BMI, social class, and employment status.

Table 5.

Final multiple regression model for New Zealand score in knees (n = 201, R² = 0.37^a).

Variableb	Category/scale	Number in category	Adjusted difference in means	95% CI	P-value
Employment status					0.049
	Paid employment (reference)	65	0
	No paid employment	10	1.7	−6.8 to 10.1
	Sick/disabled	8	8.5	−3.1 to 20.1
	Retired	118	6.1	0.9 to 11.3
Social class					0.0055
	I (reference)	11	0
	II	70	5.1	−3.4 to 13.6
	IIINM	33	−4.6	−14.5 to 5.2
	IIIM	44	3.5	−5.7 to 12.6
	IV	29	4.2	−6.2 to 14.6
	V	14	8.3	−1.8 to 18.3
Body mass index (kg/m2)			0.47c	0.01 to 0.94	0.047
Hypertension					0.041
	No (reference)	157	0
	Yes	44	5.5	0.2 to 10.8
‘Other health problems’d					0.038
	No (reference)	93	0
	Yes	108	4.6	0.3 to 8.8

^a0.25 of this is attributable to baseline, sex and age.

^bAdjusted as before for the sampling design (stratification by sex and age; clustering by general practice) as well as all the other factors.

^cThe regression coefficient is the adjusted difference in New Zealand score for a unit increase in body mass index (mean = 27.5, standard deviation = 4.6, range = 18.0 to 49.3 kg/m2).

^dOther than those specified in the list given in the methods section.

Eighty-two (28%) responders stated that they had consulted a GP about their knee problems between baseline and follow-up, and this group had deteriorated more than the others (P = 0.014). The 38 (13%) who stated that they had consulted a hospital doctor between baseline and follow-up were not worse off (difference = 1.7, 95% CI = −4.2 to 7.6; P = 0.57). Adjusting the five variables in Table 5 for having seen a GP (itself significant with an adjusted difference = 6.1, 95% CI = 1.1 to 11.1; P = 0.017) slightly attenuated the association for ‘other health problems’ (P = 0.056). It had virtually no effect on the other four variables (n = 229, R² = 0.37 for the model with these four and the GP consultation variable).

Sensitivity analyses

For all final models (with and without the variable for consulting a GP), adjusting for whether or not relevant surgery had occurred after baseline had virtually no effect on the results. Substitution of a follow-up NZ score reflecting severe joint disease (44 out of 80) similarly had very little impact on the results for knee problems. For hips the association for ‘other health problems’ was slightly weaker in this sensitivity analysis than that depicted in Table 3 (adjusted difference = 4.0, P = 0.074), as was to a similar degree the association for cardiovascular morbidity in the model that also included the GP consultation variable (results not shown).

DISCUSSION

Summary of main findings

As expected, those who reported greater knee or hip pain between 1992 and 1994 had overall more pain and worst function 7 years later. As 35% and 29%, respectively, of those initially reporting hip and knee pain had improved, however, OA does not invariably deteriorate, even though the vast majority of these individuals had not had surgical intervention. Reporting ‘other health problems’ was associated with greater deterioration for both hip and knee disease, as was cardiovascular morbidity for hip disease, and lower social class, hypertension, and higher BMI for knee disease. Our data also indicate that the NZ score is responsive to change in OA over time, with deteriorations in score strongly associated with individuals consulting their GP about joint problems independently of sociodemographic and morbidity variables. In both analyses, repeating the caveat that this might be an intermediary variable, additional adjustment for consulting a GP about the joint problem had little effect on the above relationships, other than some attenuation of the ‘other health problems’ factor. Another general observation is that the initial associations between (self-reported) depression and deteriorations in hip and knee NZ score appear to be secondary to physical comorbidities — in particular, hypertension or cardiovascular morbidity and ‘other health problems’.

Strengths and limitations of the study

One of the main strengths of this study is that it is a large community-based prospective cohort, containing a wide range of baseline data on sociodemographic, comorbidity, and healthcare utilisation variables that could be investigated in relation to change over time. Reporting was based on pain in the hip or knee, rather than a clinical or radiographic diagnosis of OA; however, as expected,^3,16 examination of the radiographs has shown that the majority of those with hip or knee pain do have OA (data available on request).

Given the aim of identifying simple predictors of change, a limitation is that our approach would not be expected to focus on potentially modifiable factors, which would have required a different design and follow-up schedule. The amount of change observed was extremely variable, but large differences in NZ scores (in either direction) were observed in many people; this suggests, as expected,⁴ that over 7 years they experienced deterioration or improvement in pain and function of major clinical significance. The optimum general approach to handling the NZ scores of those who received surgery between baseline and follow-up remains unclear, but in our case the numbers were relatively small and the results of the sensitivity analyses were reassuring.

The response rate of 66% to the 7-year follow-up questionnaire was reasonable. The youngest and oldest participants were less likely to respond to the survey, perhaps because younger people had more commitments, such as employment, and older people more ill-health and caring responsibilities. The lower response from those with eye disease may have been due to the initial contact for the survey being by letter; such individuals may need extra help with tasks like these. The poorer response among those in lower social classes is consistent with other studies.¹⁷ These differences are unlikely to have greatly influenced our main analyses.

Comparison with existing literature

There is very little prospective information on knee or hip pain or OA, and most of the existing data has concentrated on changes in radiographic status rather than pain and disability. A literature review on the progression of hip OA found that just age and BMI related to outcome, other than sophisticated radiographic findings;¹⁸ articles on predictors of hip replacement only highlighted disease severity and radiographic changes.^20,21 Similarly, knee OA progression has only been related to age, BMI, and radiographic changes in the past.^5,21 One longitudinal study suggested looking at comorbidity and psychosocial factors to explain the heterogeneity of OA outcomes.⁴ We have investigated this here, making the new observation that hypertension and cardiovascular disease, as well as social deprivation and BMI appear to predict progression. The associations with ‘other health problems’ also raise the possibility that degeneration of joints is associated with age-related changes in other systems. However, it is clear that our findings should be subjected to independent validation.

Implications for future research or clinical practice

These findings suggest that it may be possible to help clinicians predict the outcome of patients who present with hip or knee pain and OA separately from complex interpretations of the radiograph. Our findings, along with other literature,^8,10,22,23 suggest that those who are relatively deprived, out of work, overweight, have hypertension/cardiovascular disease, and other comorbidities are most likely to experience progression of pain and disability and should, therefore, be given priority for specialist referral and/or for more intensive primary care intervention. In contrast, it is clear that a more positive prognosis can be given to others, as these and other data indicate that a number of those presenting with these problems improve.

However, more research is needed on the progression (and regression) and outcome of these common complaints, including the measurement of such changes by the NZ and other scores. It is clear from our results that this should be conducted within a biopsychosocial paradigm, rather than a purely biological model, given that social and biological factors are clearly interrelated.

Funding body and reference number

This study was funded by the South West Directorate of the NHS Research & Development programme (R/44/3.96)

Ethics committee and reference number

West Somerset (7.17), Southmead (106/97), Frenchay (93/23), and United Bristol Hospitals NHS Trust (E2473) Local Research Ethics Committees granted ethical approval

Competing interests

None