| |

Received: 5 February 2019    Revised: 19 March 2019    Accepted: 22 April 2019

DOI: 10.1111/jcpe.13118

S Y S T E M AT I C R E V I E W

Predictors for tooth loss in periodontitis patients: Systematic

review and meta‐analysis

Omar Helal1 | Gerd Göstemeyer1 | Joachim Krois1 | Karim Fawzy El Sayed2,3 |

Christian Graetz2  | Falk Schwendicke1

1
Department of Operative and Preventive
Dentistry, Charité University of Berlin,
Berlin, Germany
2
Clinic of Conservative Dentistry and
Periodontology, University of Kiel, Kiel,
Germany
3
Oral Medicine and Periodontology
Department, Faculty of Oral and Dental
Medicine, Cairo, Egypt
Correspondence
Falk Schwendicke, Department of Operative
and Preventive Dentistry, Charité –
Universitätsmedizin Berlin, Aßmannshauser
Str. 4‐6, 14197 Berlin, Germany.
Email: falk.schwendicke@charite.de
Abstract
Aim: A range of predictors for tooth loss in periodontitis patients have been reported.
We performed a systematic review and meta‐analysis to assess the consistency and
magnitude of any association between a total of 12 predictors and tooth loss.
Materials and Methods: Medline/Embase/Central were searched for longitudinal
studies investigating the association between predictors and tooth loss in peri‐
odontitis patients. Random‐effects meta‐analysis was performed, and study quality
assessed.
Results: Twenty studies (15,422 patients, mean follow‐up: 12 years) were included.
The mean annual tooth loss/patient was 0.12 (min./max: 0.01/0.36). Older patients
(n = 8 studies; OR: 1.05, 95% CI: 1.03–1.08/year), non‐compliant ones (n = 11; 1.51,
1.06–2.16), diabetics (n = 7; 1.80, 1.26–2.57), those with IL‐1‐polymorphism (n = 3;
1.80; 1.29–2.52) and smokers (n = 15; 1.98, 1.58–2.48) had a significantly higher risk
of tooth loss. Teeth with bone loss (n = 3; 1.04, 1.03–1.05/%), high probing pocket
depth (n = 6; 3.19, 1.70–5.98), mobility (n = 4; 3.71, 1.65–8.38) and molars (n = 4;
4.22, 2.12–8.39), especially with furcation involvement (n = 5; 2.68, 1.75–4.08) also
showed higher risks. Gender (n = 16; 0.95, 0.86–1.05) and endodontic affection
(n = 3; 3.62, 0.99–13.2) were not significantly associated with tooth loss.
Conclusions: Older, non‐compliant, smoking or diabetic patients, and teeth with
bone loss, high probing pocket depth, mobility, or molars, especially with furcation
involvement showed higher risks of tooth loss.

KEYWORDS
periodontal therapy, periodontitis, prediction, risk model, tooth loss

1 |  I NTRO D U C TI O N
Periodontitis is the sixth most prevalent disease of humankind, affect‐
ing billions of individuals and generating considerable healthcare costs
(Kassebaum et al., 2014; Listl, Galloway, Mossey, & Marcenes, 2015). In
many cases and especially if untreated or unsuccessfully treated, the
disease leads to tooth loss. Dentists need to know predictors of tooth
loss in periodontitis patients to make informed decisions. If tooth loss
is unlikely, tooth retention via active and supportive periodontal treat‐
ment (APT, SPT) is probably the most effective and cost‐effective treat‐
ment option long‐term (Schwendicke, Graetz, Stolpe, & Dorfer, 2014).
If, however, teeth are lost shortly after APT during SPT, significant
costs have been generated without long‐term gain (Graetz et al., 2013;
Schwendicke, Plaumann, Stolpe, Dorfer, & Graetz, 2016; Schwendicke,
Stolpe, Plaumann, & Graetz, 2016). Retaining teeth with poor prognosis
may also comprise the retention of other (adjacent) teeth.

J Clin Periodontol. 2019;46:699–712. wileyonlinelibrary.com/journal/jcpe   © 2019 John Wiley & Sons A/S. |  699
Published by John Wiley & Sons Ltd
 |
700       HELAL et al.
So far, many predictors for tooth loss have been reported, both
on patient level (e.g., sex, age, compliance, smoking status, IL‐1
gene polymorphism) and tooth level (e.g., bone and attachment loss
and probing depth, mobility, furcation involvement) (Chambrone,
Chambrone, Lima, & Chambrone, 2010). Some of them, like compli‐
ance or furcation involvement, seem consistently associated with
tooth loss, as evidenced by systematic syntheses (Lee, Huang, Sun,
& Karimbux, 2015; Nibali et al., 2016). Generally, however, the con‐
sistency and magnitude of any association between predictors and
tooth loss remains unclear, that is (a) if the same predictors are re‐
ported to be relevant (and/or statistically significant) across differ‐
ent studies and populations, and (b) which predictors are, overall,
more or less important.
The present study aimed to systematically assess predictors of
tooth loss in periodontitis patients and to synthesize data on the
association between different predictors and tooth loss. Our re‐
view question was as follows: In patients with periodontitis, having
received active periodontal treatment (APT) and being provided
supportive periodontal treatment (SPT) over at least 3 years, which
predictors are (or are not) associated with tooth loss?
2 |  M ATE R I A L S A N D M E TH O DS
Reporting of this review and meta‐analysis was performed accord‐
ing to established guidelines (Moher, Liberati, & Tetzlaff, 2009).
The study protocol was registered after the initial screening stage
(PROSPERO CRD42017057841). Any deviations from the protocol
are described below.
2.1 | Eligibility criteria
The following selection criteria were applied: (1) studies needed to
report on dentate humans with periodontitis which received APT
and SPT, without further specification for diagnoses (which may
have changed given that no time restriction was applied) and treat‐
ment concepts. (2) We included pro‐ and retrospective observational
studies (e.g., cohort studies or case series) as well as interventional
longitudinal studies (e.g., randomized controlled trials, as long as the
intervention was not designed to modify the association between a
predictor and tooth loss). (3) Only studies with a minimum of 3 years
follow‐up period investigating the association between predictors
(independent variables) and tooth loss (outcome) were included, as
we expected the outcome tooth loss to not occur in relevant fre‐
quency in short‐term studies, and as dentists are interested into
making longer‐term decisions. Note that eventually, most included
studies showed much longer follow‐up periods, which allows to infer
on truly long‐term associations between predictors and the outcome.
(4) Studies needed to have used a multivariable model to evaluate the
association between predictors and the outcome, accounting for a
minimum of three predictors. We did not specify how exactly tooth
loss was measured or reported, because we expected a large range of
different definitions. (5) Studies needed to have reported associations
Clinical Relevance
Scientific rationale for the study: Predictors of tooth loss in
periodontitis patients may assist dental treatment planning
and risk communication.
Principal findings: Older, non‐compliant, smoking or dia‐
betic patients, and teeth with bone loss, high probing
pocket depth, mobility or molars, especially with furcation
involvement showed higher risks of tooth loss. Tooth‐level
predictors were largely more accurate and showed a higher
of association with tooth loss than patient‐level predictors.
Practical implications: Dentists may cautiously apply a
defined set of predictors, best in combination with each
other, for clinical decision‐making. Generalizability to all
patients and settings may not be given.
between predictors and tooth loss via odds ratios, risk ratios or their
derivatives, and needed to have included an uncertainty estimate
(confidence interval, p‐value, standard error) to allow entering into
meta‐analysis (see below). Note that this may have led to exclusion
of studies potentially eligible to the review and non‐statistical, that is
narrative synthesis. We accepted that as one of the main goals of this
study was to perform statistical synthesis. Only studies fulfilling all of
the above described criteria were included.
2.2 | Information sources and search
We systematically screened three electronic databases (Medline,
Embase, Cochrane Central) for studies published up to 30th
November 2018. The search strategy combined the study focus (risk
OR prognostic OR prognosis OR prediction OR predict OR associa‐
tion OR correlation OR hazard) with the condition (periodontitis OR
periodontal disease), the outcome (tooth loss OR teeth lost) and the
population (patients OR individuals) using Boolean operators. No re‐
strictions (MeSH etc.) were applied (i.e., only free terms were used).
Unpublished studies or grey literature were excluded, because we
expected insufficient reporting for our analysis. Articles such as re‐
views or editorials were used to identify original studies, and cross‐
referencing from bibliographies performed. Only studies published
in English were included; the language restriction was a pragmatic
decision and may result in under‐detection of eligible studies.
2.3 | Screening
Screening of titles or abstracts was independently performed by
two reviewers (OH, KFE). Any disagreement was resolved by discus‐
sion or consulting a third reviewer (FS). All papers which were found
to be potentially eligible were assessed in full‐text against the inclu‐
sion criteria. Inclusion and exclusion were decided by two reviewers
in consensus (OH, KFE).
HELAL et al.
2.4 | Data collection, items and preprocessing
A pretested and standardized spreadsheet‐based data collection
form was used. Study characteristics, including study type, place of
conduct, year of publication, sample size, mean follow‐up period, case
definition for periodontitis (as aggressive or chronic periodontitis,
which was the classification most studies used), assessed predictors
(risk factors or indicators), tooth loss per year and patient (in some
cases, we calculated this from the reported data of lost teeth, sample
size and mean follow‐up period), as well as findings (adjusted associa‐
tion estimates) were extracted from the most comprehensive report
of each study independently by two reviewers (OH, GG). Study au‐
thors were contacted if data were missing or for clarifications.
A number of decisions were needed during extraction. If one
study reported on associations between tooth loss and multiple di‐
mensions of the same independent variable (e.g., smoking status),
we extracted the estimate capturing the largest difference between
categories in order to explore the maximal extent of any association
(Conway et al., 2007). If several models were reported on, adjusted
estimates were extracted from the model that included the largest
number of relevant predictors (Higgins & Green, 2011). If the same
survey/study was reported on in multiple articles, the article with the
largest sample size was included in order to avoid unit‐of‐analysis is‐
sues. In one case, two reports had a minimal overlap of six patients,
and we decided to nevertheless include both reports to not unneces‐
sarily loose data, accepting this minimal overlap (Baumer et al., 2011;
Eickholz, Kaltschmitt, Berbig, Reitmeir, & Pretzl, 2008).
Association estimates were transformed if needed, in order to pro‐
vide identically directed input data for meta‐analysis. Similarly, avail‐
able uncertainty estimates were transformed into standard errors for
meta‐analysis (Altman & Bland, 2003; Higgins & Green, 2011). Effect
estimates reported as “statistically non‐significant” without any fur‐
ther numerical values presented were imputed as recommended by
the Cochrane collaboration (Higgins & Green, 2011), assuming the
estimate to be “1” and the Standard Error (SE) being the mean of the
reported SEs for each analysis. If studies reported separately on dif‐
ferent groups of participants from the same study, data were pooled
for meta‐analysis using random‐effect meta‐analysis before entering
them into our main meta‐analysis (Higgins & Green, 2011).
For meta‐analysis, independent variables (i.e., predictors), which
had been reported in various formats and using various measures,
were pooled into the following categories: on patient level (1) sex
(male versus female), (2) age (we only pooled studies which reported
on the association with tooth loss per year of age, as this was the
majority of studies), (3) compliance (compliant versus non‐compliant;
some studies measured this via the number of annual SPT visits in a
dichotomized fashion), (4) smoking status (usually comparing current
smokers versus never/former smokers), (5) diabetes mellitus (present
versus not), (6) IL‐1‐polymorphism (being present or not). On tooth
level, (7) endodontic conditions present (which included root fillings
or periapical lesions being present versus not); (8) tooth type (we only
used molars or multi‐rooted teeth versus other); (9) maximal probing
pocket depth (PPD; we only pooled dichotomized comparisons, that
|
      701
is higher PPD versus lower PPD, as only one study reported the as‐
sociation of tooth loss per mm PPD); (10) furcation involvement (FI;
we only pooled the worst degree versus none or the mildest degree);
(11) bone loss (BL; we only pooled the studies which reported on the
association with tooth loss per % BL, as this was the majority of stud‐
ies); (12) mobility (again, we pooled the worst degree versus none or
the mildest degree). Note that studies also reported on further pre‐
dictors (e.g., bleeding on probing, various oral hygiene indices, tooth
abutment status, follow‐up period). As all of these were reported in
fewer than three studies, we did not consider them further here, as
such sparse data are unlikely to yield robust information.
Most studies reported estimates using odds ratios (OR); only a mi‐
nority (6 of the total 20 included studies, see below) used the risk or
hazard ratios (RR or HR). We entered all of them into one meta‐analy‐
sis and performed subgroup analyses for studies using OR versus RR/
HR to gauge the impact of pooling different estimate types. As tooth
loss is generally a rather infrequent event, only minimal differences
between OR and RR/HR were expected (Higgins & Green, 2011).
2.5 | Quality assessment
As there is currently no validated evaluation tool available for qual‐
ity assessment in both cohort studies and case series, an evaluation
instrument originally developed for risk of bias assessment in case
reports and case series was applied (Bazerbachi, Haffar, et al., 2017;
Bazerbachi, Leise, et al., 2017; Haffar et al., 2017). This tool is based
on the Newcastle–Ottawa Scale (NOS), which is accepted standard
for assessing non‐randomized studies (Wells et al., 2009). By remov‐
ing items from the original NOS which are not relevant for non‐com‐
parative study types, this tool comprised a 5‐item questionnaire (1.
Did the patient(s) represent the whole case(s) of the medical centre?
2. Was the diagnosis correctly made? 3. Were other important di‐
agnosis excluded? 4. Were all important data cited in the report?
5. Was the outcome correctly ascertained?) with a binary response
option (yes/no) for each item. Studies were considered as having a
high methodical quality if all, moderate if 4 and low quality if 3 or
less questions were answered affirmatively. Quality assessment was
performed independently by two reviewers (OH, GG).
2.6 | Summary measures and data synthesis
Data preparation, meta‐analysis and visualization were performed
using the statistical programming language R, version 3.5.1, and in
particular the metafor package, an open‐source toolbox for conduct‐
ing meta‐analyses (Viechtbauer, 2010). The summary measure was
OR and 95% CI. Standard errors were used to estimate 95% CI. Note
that this, together with numerical rounding, may introduce minimal
differences between the published CI and those used for our analyses.
Overall, twelve subgroup analyses (one for each of the twelve inde‐
pendent variables) were performed. Heterogeneity was assessed using
Cochrane's Q and I2 statistics (Higgins & Thompson, 2002). Since het‐
erogeneity was usually found to be substantial (for two analyses I2 was
>90% and for six analyses I2 was >70%), random‐effect models were
 |
702       HELAL et al.
used. Publication bias or selective reporting was evaluated using funnel
plots. If five or more studies were present, Egger's regression intercept
test was also performed (Egger, Smith, Schneider, & Minder, 1997).
Meta‐regression was applied to examine the contribution of
the characteristics of the studies to the heterogeneity of treatment
effects across the studies. In meta‐regression, the effect sizes cor‐
respond to the outcomes and the study characteristics to the pre‐
dictors. We performed the regression analysis using the metafor
package and computed the random‐effects variance component by
restricted maximum likelihood (Thompson & Sharp, 1999).
3 |   R E S U LT S
3.1 | Study selection and characteristics
Our search yielded 3,474 records, with 140 articles being possibly
eligible after review on abstract level (Figure 1). After full‐text re‐
view, 120 studies were excluded (Table S1) and 20 studies included
(Table 1). The mean (min./max.) year of publication of the included
studies was 2011 (2004/2016). The mean follow‐up period was
12 years (range of study means: 4–20). The studies involved a total
of 15,422 participants. The mean (min./max.) sample size was 771
(25/12,631). Patients’ mean age was 44.5 (range of study means:
31–54) years. Eight studies assessed only patients with chronic,
three with aggressive periodontitis; the remaining studies either as‐
sessed both types of disease or did not provide information on this
issue. Mean annual tooth loss per patient was 0.12 (430.01/0.36).
Studies had been performed in 12 countries; the most frequent were
Germany (5 studies), Brazil (2), Switzerland (2), Italy (2) and Spain (2).
Five of the 20 studies had been performed in general practice, the
remaining ones in specialized care. Two studies recruited a selected
patient pool (patients attending an army hospital and healthcare sys‐
tem employees) with periodontitis (Cunha‐Cruz, Hujoel, Maupome,
& Saver, 2008; Muller, Eickholz, Reitmeir, & Eger, 2013). Two studies
only assessed molars (Baumer et al., 2011; Graetz et al., 2015).
3.2 | Risk of bias
Most studies (9) had a moderate, 5 studies had a high and 6 studies
had a low methodological quality (Table S2). As described, we fur‐
ther explored the risk of bias stemming from pooling studies which
reported effects as OR and RR/HR by performing subgroup analy‐
ses. The resulting changes in the pooled estimate when separately
analysing these groups were limited; only for diabetes and smoking,
the pooled estimate differed significantly between the subgroups
(p < 0.05), with studies using OR yielding higher pooled estimates.
3.3 | Associations with patient‐level parameters
Based on 3 to 16 studies, the associations of tooth loss with six dif‐
ferent patient‐level parameters were explored (Figure 2).
• Age (n = 8 studies) was found significantly associated with tooth
loss. With each year of age, the odds of tooth loss increased by
F I G U R E 1   Flowchart of the search
TA B L E 1   Included studies

Mean Teeth Tooth Tooth Specialized

Mean CP/ Sample follow‐up remaining loss in 3rd loss per care or general Predictor Comparator Reference Outcome
HELAL et al.

Study Year Country age AP size (years) after APT SPT molars pt./year practice variable (worst) (best) measure

Bäumer 2011 2011 Germany 31 AP 84 11 2,154 113 Excluded 0.13 Specialized Gender Male Female OR
(Baumer et al., Smoking Current Never OR
2011) smoker smoked
Furcation Yes No OR
involvement
Molar Molar Premolar OR
Carollo‐Bittel 2011 2011 Switzerland ‐ CP 89 10 na na Excluded 0.10 Specialized Smoking Smoker Non‐smoker OR
(Carollo‐Bittel, PD ≥6 mm <6 mm OR
Persson, Persson,
& Lang, 2011) IL‐1 Positive Negative OR

Chambrone 2006 2006 Brazil 39 CP 120 17 2,927 111 Included 0.05 Specialized Gender Male Female OR
(Chambrone Smoking Smoker Non‐smoker OR
& Chambrone,
2006) Compliance Once/year Twice/year OR

Costa 2014 (Costa 2014 Brazil ‐ CP 212 5 4,887 234 Mixed 0.22 General Compliance Irregular Regular OR
et al., 2014) Gender Male Female OR
Smoking Smoker Non‐smoker OR
Diabetes Yes No OR
PD 4–6 mm up <4 mm or OR
to 10% of <10% sites
sites
Cunha‐Cruz 2008 2008 USA 54 CP 12,631 4 na 6,681 na 0.1 General Gender Male Female RR
(Cunha‐Cruz Diabetes Yes No RR
et al., 2008)
Smoking Smoker Never RR
smoker
Di Febo 2015 (Di 2015 Italy 42 na 100 20 948 94 na 0.05 General Age/year – – OR
Febo, Bedendo, Gender Male Female OR
Romano, Cairo, &
Carnevale, 2015)
Díaz‐Faes 2016 2016 Spain 31 AP 25 11 656 28 Excluded 0.11 Specialized Age/year – – OR
(Diaz‐Faes, Gender Male Female OR
Guerrero, Magan‐
Fernandez, Bravo, Smoking Smoker Non‐smoker OR
& Mesa, 2016) Compliance Irregular Regular OR
|
      703

(Continues)
 |

TA B L E 1   (Continued)
704      

Mean Teeth Tooth Tooth Specialized

Mean CP/ Sample follow‐up remaining loss in 3rd loss per care or general Predictor Comparator Reference Outcome
Study Year Country age AP size (years) after APT SPT molars pt./year practice variable (worst) (best) measure

Eickholz 2008 2008 Germany 47 – 100 10 2,246 155 Excluded 0.16 Specialized Compliance Non‐com‐ Compliant RR
(Eickholz et al., pliant
2008) IL‐1 Positive Negative RR
Smoking Smoker Former/ RR
never
Gender Male Female RR
Age/year – – RR
Diabetes Yes No RR
Faggion 2007 2007 Germany 47 – 198 12 4,393 249 Excluded 0.11 Specializes Diabetes Yes No OR
(Faggion, BL in % – – OR
Petersilka, Lange,
Gerss, & Flemmig, Mobility III None OR
2007) Molar Multi‐rooted Single OR
rooted
Endodontic Non‐vital Vital OR
treatment
Fardal 2004 2004 Norway 46 CP 100 10 2,436 36 Included 0.04 Specialized Gender Male Female OR
(Fardal, Compliance Once/year Twice/year OR
Johannessen, &
Linden, 2004) Smoking Smoker Non‐smoker OR

Graetz 2015 2015 Germany 46 Both 379 20 1,892 357 Excluded 0.05 Specialized Furcation III None HR
(Graetz et al., involvement
2015) Mobility Any None HR
Endodontic Yes No HR
treatment
Age/year – – HR
Gender Male Female
Smoking Current Never/former
Diabetes Yes No
Leung 2006 2006 China 41 AP 97 9 2,522 256 Included 0.25 Specialized Age/year – – OR
(Leung, Ng, Jin, & Gender Male Female
Corbet, 2006)

(Continues)
HELAL et al.
 HELAL et al.

TA B L E 1   (Continued)

Mean Teeth Tooth Tooth Specialized

Mean CP/ Sample follow‐up remaining loss in 3rd loss per care or general Predictor Comparator Reference Outcome
Study Year Country age AP size (years) after APT SPT molars pt./year practice variable (worst) (best) measure

Martinez‐Canut, 2015 Spain 40 Both 500 20 12,595 640 Excluded 0.05 Specialized Smoking Heavy Non‐smoker RR
2015; (Martinez‐ smoker
Canut, 2015) Compliance Irregular Regular RR
Age/year – – RR
Gender Male Female RR
Mobility III 0 RR
(molars)
Furcation in‐ III 0 RR
volvement
(molars)
PD (molars) >6 mm Max. 6 mm RR
Mobility III 0 RR
(non‐molars)
PD >6 mm Max. 6 mm RR
(non‐molars)
Muller 2013 2013 Germany 48 Both 90 10 2,230 387 Excluded 0.36 Specialized Gender Male Female RR
(Muller et al., Age/year – – RR
2013)
Smoking Smoker Non‐smoker RR
Diabetes Yes No RR
BL in % – – RR
Compliance Partially Full RR
Muzzi 2006 (Muzzi 2006 Italy 47 CP 60 13 1,495 na Excluded na Specialized Gender Male Female OR
et al., 2006) IL‐1 Positive Negative OR
Mobility Yes No OR
Molar Yes No OR
Ng 2011 (Ng, Ong, 2011 Singapore 45 CP 273 11 6,726 253 Included 0.09 Specialized Diabetes Yes No OR
Lim, Koh, & Chan, Gender Male Female OR
2011)
Compliance Regular Irregular OR

(Continues)
|
      705
 |
706      

TA B L E 1   (Continued)

Mean Teeth Tooth Tooth Specialized

Mean CP/ Sample follow‐up remaining loss in 3rd loss per care or general Predictor Comparator Reference Outcome
Study Year Country age AP size (years) after APT SPT molars pt./year practice variable (worst) (best) measure

Nibali 2016 (Nibali 2016 UK 53 CP 98 8 2,395 45 Excluded 0.06 General Gender Male Female OR
et al., 2016) Smoking Smoker Non‐smoker OR
Age/year – – OR
Furcation Yes No OR
involvement
Endodontic Yes No OR
treatment
BL in % – – OR
Ravald & 2012 Sweden 52 na 64 13 1,537 211 na 0.23 General Compliance <1 visit/ Once or OR
Johansson, year twice/year
2012; (Ravald & Smoking Smokers Non‐smok‐ OR
Johansson, 2012) ers
PD >6 4–6 OR
Salvi 2014 (Salvi 2014 Switzerland na Both 152 12 113 17 Excluded 0.01 Specialized Furcation III 0 OR
et al., 2014) involvement
Smoking Smoker Non‐smoker OR
Compliance Non‐com‐ Compliant OR
pliant
PD ≥6 mm <6 mm OR
Gender Male Female OR
Saminsky 2015 2015 Israel 47 na 50 13 1,281 129 na 0.16 Specialized Molar Multi‐rooted Single HR
(Saminsky, rooted
Halperin‐ Compliance 2–3 times/ ≥3 times/ HR
Sternfeld, year year
Machtei, &
Horwitz, 2015) PD ≥7 mm 1–3 mm HR
Smoking Smoker Non‐smoker HR

Abbreviations: AP/CP, aggressive/chronic periodontitis; HR, hazard ratio; OR, odds ratio; PD, pocket probing depths; RR, risk ratio.
HELAL et al.
HELAL et al.
5% in mean (OR: 1.05, 95% CI: 1.03–1.08), with only limited
heterogeneity between studies. There was indication for pub‐
lication bias (Egger test p < 0.01, see funnel plot in Figure S1).
• Patients’ compliance (n = 11) was also significantly associ‐
ated; patients with lower compliance showed significantly in‐
creased odds of tooth loss (1.51, 1.06–2.16). There was great
heterogeneity, but no indication for publication bias (p > 0.05,
Figure S1).
• Diabetes (n = 7) was also significantly increasing the odds of tooth
loss (1.80, 1.26–2.57), again with great heterogeneity and signifi‐
cant risk of publication bias (p < 0.01, Figure S1).
• Gender (n = 16) did not show a significant association with tooth
loss (0.95, 0.86–1.05). There was no indication for publication bias
(p &gt; 0.05, Figure S1).
• IL‐1 gene polymorphism (n = 3) was significantly associated with
tooth loss (1.80; 1.29‐2.52).
• Smoking (n = 15) was significantly associated with tooth loss (1.98,
1.58–2.48), with great heterogeneity and a risk of publication bias
(p = 0.03, Figure S1).
3.4 | Associations with tooth‐level parameters
Based on 3 to 6 studies, the associations of tooth loss with six differ‐
ent tooth‐level parameters were explored (Figure 3).
• BL (n = 3) significantly increased the odds of tooth loss; with each
% of BL, the odds of tooth loss increased by 1.04 (1.03–1.05).
There was low heterogeneity.
• An existing endodontic treatment or apical lesion (n = 3) had
no significant impact on tooth loss (3.62, 0.99–13.2), with high
heterogeneity.
• Furcation involvement (n = 5) significantly increased the odds of
tooth loss (2.68, 1.75–4.08). There was limited indication for pub‐
lication bias (p > 0.05, Figure S1).
• Mobility (n = 4) was significantly associated with tooth loss (3.71,
1.65–8.38), again with high heterogeneity. There was no indica‐
tion for publication bias (p > 0.05, Figure S1).
• Molars (n = 4) had significantly increased odds of tooth loss (4.22,
2.12–8.39), with great heterogeneity. There was no indication for
publication bias (p > 0.05, Figure S1).
• PPD (n = 6) was significantly associated with the odds of tooth
loss (3.19, 1.70–5.98). There was indication for publication bias
(p < 0.01, Figure S1).
3.5 | Meta‐regression
Table 2 displays the results of meta‐regression analysis. On patient
level, compliance (1.52; 1.10–2.11), diabetes (1.84; 1.23–2.73) and
smoking (1.91; 1.54–2.78) were significantly associated with tooth
loss. On tooth level, all factors were significantly associated except
BL; the strongest associations were found for tooth type (in mean,
molars showed 4‐times increased risk of loss) and mobility (3.5‐times).
|
      707
4 | D I S CU S S I O N
Predicting periodontitis onset, progression and the tooth loss re‐
sulting from periodontitis is relevant for dentists during treatment
planning, but also risk communication with patients. Prediction
models for periodontitis incidence and progression have been sys‐
tematically evaluated recently (Du, Bo, Kapellas, & Peres, 2018).
No such evaluation for predictors of tooth loss is available, and
while a range of tooth loss prediction systems have been estab‐
lished, most have not at all or only limitedly been externally vali‐
dated (Schwendicke et al., 2018). We systematically reviewed and
meta‐analysed (observational) studies on tooth loss in periodontitis
patients, allowing to identify both ambiguity but also agreement
between studies.
This review has a number of strengths and limitations. First,
data collection was performed systematically, yielding a total of
20 studies with over 15,000 participants with periodontitis. Many
of these studies had a long‐term follow‐up (mean follow‐up period
was 12 years, which is remarkable), and investigated a large range
of risk factors assessed. Consequently, many of our meta‐analyses
were supported by five or more studies, which increases their sta‐
tistical power and robustness. Second, our outcome, tooth loss, is
a relatively “hard” endpoint and not easy to bias; notably, though,
not all tooth losses recorded by the included studies were due
to periodontal reasons (but also endodontic or restorative ones).
Third, and as a limitation, none of the studies were truly represen‐
tative regarding the provided care and for all populations; instead,
most studies assessed a selected population of largely compliant
patients with long‐term SPT provided in a specialized setting. Also,
the proportion of sampled participants with aggressive periodonti‐
tis was relatively high. Considering that annual tooth loss rates per
patient have not necessarily been found to differ between chronic
and aggressive periodontitis patients, this may be of limited rele‐
vance (Graetz et al., 2017; Nibali, Farias, Vajgel, Tu, & Donos, 2013;
Schwendicke, Biffar, & Graetz, 2017). Fourth, a high heterogeneity
in sample characteristics (which increases generalizability but also
statistical noise), treatment concepts, and definition and reporting
of predictors was present. This has likely decreased the power of
our analyses; for example, we found homogenous predictor defini‐
tions (e.g., age per year, or BL in %) to come with much lower sta‐
tistical heterogeneity and narrower confidence intervals compared
with other, more variable predictor definitions (where cut‐off for
categories or, generally, category definition was not always uni‐
form). On the other hand, this heterogeneity allows to contrast the
findings from different settings and populations, too, and some of
it might indeed be associated with clinical variability. For example,
for a range of predictors, some studies found very strong associ‐
ations, while others found none at all (this was most obvious for
IL‐1 gene polymorphism), indicating either a variability in the true
association (different IL‐1 gene polymorphisms are known and have
different effects in different periodontitis phenotypes and ethnic
groups) (Karimbux et al., 2012) or a variability in measurements of
the predictor. We also were unable to identify the properties of
 |
708       HELAL et al.

F I G U R E 2   Forest plot of studies on the association between patient‐level parameters and tooth loss. Squares indicate the estimates
(Odds, Risk or Hazard Ratio) of single studies; lines the 95% confidence intervals (95% CI). Black diamonds show the pooled estimates. Q and
I2 indicate heterogeneity
HELAL et al. |
      709

F I G U R E 3   Forest plot of studies on the association between tooth‐level parameters and tooth loss. Squares indicate the estimates
(Odds, Risk or Hazard Ratio) of single studies; lines the 95% confidence intervals (95% CI). Black diamonds show the pooled estimates. Q and
I2 indicate heterogeneity
 |
710       HELAL et al.

TA B L E 2   Meta‐regression analysis tooth loss in the included studies (e.g., some studies only sampled
molars) but also the performed treatment concept (which we did
Factor OR LCI UCI
not consider further in this review). Second, certain predictors were
Age/year 1.07 0.80 1.43
consistently associated with tooth loss. Older patients, smokers and
Compliance (ref.: 1.52 1.10 2.11 diabetics suffered from more tooth loss. The last two predictors
compliant)
are causally associated with the patho‐mechanism of periodonti‐
Diabetes (ref.: no) 1.84 1.23 2.73
tis. Compliance was also a significant predictor, but in contrast to
Gender (ref.: female) 0.98 0.75 1.27
the other three, this was not uniformly the case across studies. As
IL‐1 polymorphism (ref.: 1.67 0.79 3.54 compliance may serve as an indicator of further, uncaptured pa‐
no)
tient characteristics, the magnitude of any association will be con‐
Smoking (ref.: none/ 1.91 1.54 2.78
founded and will be larger for models where only few confounders
former)
were included (Lee et al., 2015). BL and PPD were also found to be
Bone loss in % 1.03 0.64 1.66
associated, being indicators of disease activity and past disease ex‐
Endodontic treatment 3.10 1.73 5.56
perience. Similarly, mobility serves as an indicator of disease expe‐
(ref.: no)
rience and was significantly associated with tooth loss. BL and PPD
Furcation involvement 2.73 1.68 4.43
are also considered by the recently introduced new classification
(ref.: lower)
of periodontal diseases (Papapanou et al., 2018). For mobility, the
Mobility (ref.: lower) 3.49 1.95 6.24
reliability of determining the exact degree has been found to be
Molar (ref.: no) 4.05 2.35 6.98
limited (Laster, Laudenbach, & Stoller, 1975), and in some included
Probing pocket depths 3.08 2.04 4.6
studies, mobility measurements were further distorted by tooth
(ref.: lower)
splinting (Graetz et al., 2015). Molars, especially those with furca‐
Note: The association estimates and lower/upper confidence intervals
tion involvement, were at higher risk, too. Multi‐rooted teeth have
(LCI/UCI) are provided. Bold indicates significant associations.
been reported to be generally more prone for periodontal disease
and to be more challenging to manage during APT and SPT. Molars
any associations; for example, it is conceivable that associations are also more often heavily restored and/or endodontically treated,
are not always linear, but exponential. Also, any kind of synergism which may lead to tooth removal due to non‐periodontal reasons in
between predictors (e.g., multiplication of risk when smoking and some cases (which, as discussed, was often not separately reported).
non‐compliance come together) could not be further explored. However, even molars can be retained long‐term at limited costs
Last, we included only studies in English. This may explain some (Dannewitz et al., 2016; Graetz et al., 2015; Schwendicke, Graetz,
of the identified publication bias. However, based on our statistical Stolpe, & Dorfer, 2014).
assessment and inspection of funnel plots, we assume this to have A number of predictors were not significantly or consistently
had only limited impact. Also, the vast majority of included stud‐ associated with tooth loss in our meta‐analyses. The presence of a
ies stemmed from high‐income countries and all patients were in‐ root‐filling or apical lesion was found to be a significant predictor
cluded in SPT. Associations of predictors with tooth loss may differ by all three studies reporting on this factor, but because of het‐
in patients not attending SPT, and annual tooth loss per patient will erogeneity, random‐effects meta‐analysis yielded a non‐significant
differ between settings and populations (Lee et al., 2015). pooled estimated. Hence, caution is needed when interpreting this
A number of decisions were needed during this study to allow data non‐significance, as it is a result of uncertainty and not necessary
synthesis. For example, we could only include data from certain data of a lacking association. Generally, the loss of pulp vitality and the
points in case the same cohort was analysed in several reports. While presence of a root‐canal filling or apical lesion has been found to
the decision which data point to include was defined a priori, using increase the risk of tooth loss, not only in periodontitis patients
different points may have yielded different findings (at least to some (Ng, Mann, & Gulabivala, 2010). After accounting for effects being
degree). Pooling only the highest association estimates was useful for introduced by the study itself via meta‐regression, such associa‐
our study to capture the possible extent of an association, but again tion between tooth loss and endodontic status was confirmed by
depends on the described categorization in the included studies. our study, too. Gender was not significantly associated; men and
Some categorization may, by design, not capture differences as well as women seem at similarly higher risk of tooth loss. Also, IL‐1 gene
others. We were also not able to describe the nature of any gradient in polymorphism was not found consistently associated with tooth
the associations (except for those meta‐analyses where the predictors loss (while there was generally only sparse data on this predictor).
were entered on a continuous scale, like BL or age), as described. As discussed, the ambiguity of study findings was extreme here,
A number of aspects identified by our study need discussion. with one study finding very strong (albeit non‐significant) associa‐
First, the mean annual tooth loss rate per patient was low at 0.12 tion, one finding none at all and one finding only a moderate one.
teeth. Over 10 years of SPT, each periodontitis patient will lose, in Given the high efforts and costs for recording this predictor, a rou‐
mean, only 1 tooth. However, there was again great heterogeneity tine use may not be justified based on this review and the therein
across studies, which likely stemmed from different baseline risk of included studies (Higashi, Veenstra, del Aguila, & Hujoel, 2002).
HELAL et al.
5 | CO N C LU S I O N
Based on this review and meta‐analysis, older, non‐compliant pa‐
tients, smokers, diabetics, those with IL‐1‐polymorphism, and teeth
with bone loss, high probing pocket depths, mobility, as well as
molars, especially with furcation involvement, are at higher risk of
tooth loss. In conclusion, certain, mainly tooth‐level predictors were
consistently found associated with tooth loss. However, given the
high heterogeneity across studies, generalizability to all patients and
settings may not be given. Dentists should cautiously apply predic‐
tors for the risk of tooth loss during clinical decision‐making.
AC K N OW L E D G E M E N T S
This study was funded by the authors and their institutions.
C O N FL I C T O F I N T E R E S T
The authors declare no conflict of interest.
ORCID
Christian Graetz  https://orcid.org/0000-0002-8316-0565
Falk Schwendicke  https://orcid.org/0000-0003-1223-1669
REFERENCES
Altman, D. G., & Bland, J. M. (2003). Interaction revisited: The difference
between two estimates. BMJ, 326, 219. https​ ://doi.org/10.1136/
bmj.326.7382.219
Baumer, A., Pretzl, B., Cosgarea, R., Kim, T. S., Reitmeir, P., Eickholz, P.,
& Dannewitz, B. (2011). Tooth loss in aggressive periodontitis after
active periodontal therapy: Patient‐related and tooth‐related prog‐
nostic factors. Journal of Clinical Periodontology, 38, 644–651. https​
://doi.org/10.1111/j.1600-051X.2011.01733.x
Bazerbachi, F., Haffar, S., Szarka, L. A., Wang, Z., Prokop, L. J., Murad, M.
H., & Camilleri, M. (2017). Secretory diarrhea and hypokalemia asso‐
ciated with colonic pseudo‐obstruction: A case study and systematic
analysis of the literature. Neurogastroenterology and Motility, 29, https​
://doi.org/10.1111/nmo.13120​
Bazerbachi, F., Leise, M. D., Watt, K. D., Murad, M. H., Prokop, L. J., & Haffar,
S. (2017). Systematic review of mixed cryoglobulinemia associated with
hepatitis E virus infection: Association or causation? Gastroenterology
Report, 5, 178–184. https​://doi.org/10.1093/gastr​o/gox021
Carollo‐Bittel, B., Persson, R. E., Persson, G. R., & Lang, N. P. (2011).
Supportive periodontal therapy for high‐ and low‐risk patients.
Journal of Investigative and Clinical Dentistry, 2, 110–116. https​://doi.
org/10.1111/j.2041-1626.2011.00050.x
Chambrone, L. A., & Chambrone, L. (2006). Tooth loss in well‐maintained
patients with chronic periodontitis during long‐term supportive ther‐
apy in Brazil. Journal of Clinical Periodontology, 33, 759–764. https​://
doi.org/10.1111/j.1600-051X.2006.00972.x
Chambrone, L., Chambrone, D., Lima, L., & Chambrone, L. (2010).
Predictors of tooth loss during long‐term periodontal maintenance:
A systematic review of observational studies. Journal of Clinical
Periodontology, 37, 675–684.
Conway, D. I., Quarrell, I., McCall, D. R., Gilmour, H., Bedi, R., & Macpherson,
L. M. (2007). Dental caries in 5‐year‐old children attending multi‐ethnic
|
      711
schools in Greater Glasgow–the impact of ethnic background and levels
of deprivation. Community Dental Health, 24, 161–165.
Costa, F. O., Lages, E. J., Cota, L. O., Lorentz, T. C., Soares, R. V., & Cortelli,
J. R. (2014). Tooth loss in individuals under periodontal maintenance
therapy: 5‐year prospective study. Journal of Periodontal Research,
49, 121–128. https​://doi.org/10.1111/jre.12087​
Cunha‐Cruz, J., Hujoel, P., Maupome, G., & Saver, B. (2008). Systematic
antibiotics and tooth loss in periodontal disease. Journal of Dental
Research, 87, 871–876.
Dannewitz, B., Zeidler, A., Husing, J., Saure, D., Pfefferle, T., Eickholz,
P., & Pretzl, B. (2016). Loss of molars in periodontally treated pa‐
tients: Results 10 years and more after active periodontal therapy.
Journal of Clinical Periodontology, 43, 53–62. https​://doi.org/10.1111/
jcpe.12488​
Di Febo, G., Bedendo, A., Romano, F., Cairo, F., & Carnevale, G. (2015).
Fixed prosthodontic treatment outcomes in the long‐term manage‐
ment of patients with periodontal disease: A 20‐year follow‐up re‐
port. International Journal of Prosthodontics, 28, 246–251. https​://doi.
org/10.11607/​ijp.3995
Diaz‐Faes, L., Guerrero, A., Magan‐Fernandez, A., Bravo, M., & Mesa, F.
(2016). Tooth loss and alveolar bone crest loss during supportive peri‐
odontal therapy in patients with generalized aggressive periodonti‐
tis: Retrospective study with follow‐up of 8 to 15 years. Journal of
Clinical Periodontology, 43, 1109–1115. https​ ://doi.org/10.1111/
jcpe.12596​
Du, M., Bo, T., Kapellas, K., & Peres, M. A. (2018). Prediction models
for the incidence and progression of periodontitis: A systematic re‐
view. Journal of Clinical Periodontology, 45, 1408–1420. https​://doi.
org/10.1111/jcpe.13037​
Egger, M., Smith, G. D., Schneider, M., & Minder, C. (1997). Bias in meta‐
analysis detected by a simple, graphical test. BMJ, 315, 629–634.
https​://doi.org/10.1136/bmj.315.7109.629
Eickholz, P., Kaltschmitt, J., Berbig, J., Reitmeir, P., & Pretzl, B. (2008). Tooth
loss after active periodontal therapy. 1: Patient‐related factors for risk,
prognosis, and quality of outcome. Journal of Clinical Periodontology,
35, 165–174. https​://doi.org/10.1111/j.1600-051X.2007.01184.x
Faggion, C. M. Jr, Petersilka, G., Lange, D. E., Gerss, J., & Flemmig, T. F.
(2007). Prognostic model for tooth survival in patients treated for
periodontitis. Journal of Clinical Periodontology, 34, 226–231. https​://
doi.org/10.1111/j.1600-051X.2006.01045.x
Fardal, O., Johannessen, A. C., & Linden, G. J. (2004). Tooth loss during
maintenance following periodontal treatment in a periodontal prac‐
tice in Norway. Journal of Clinical Periodontology, 31, 550–555. https​
://doi.org/10.1111/j.1600-051X.2004.00519.x
Graetz, C., Plaumann, A., Schlattmann, P., Kahl, M., Springer, C., Salzer, S.,
… Schwendicke, F. (2017). Long‐term tooth retention in chronic peri‐
odontitis ‐ results after 18 years of a conservative periodontal treat‐
ment regimen in a university setting. Journal of Clinical Periodontology,
44, 169–177. https​://doi.org/10.1111/jcpe.12680​
Graetz, C., Schutzhold, S., Plaumann, A., Kahl, M., Springer, C., Salzer,
S., … Schwendicke, F. (2015). Prognostic factors for the loss of
molars–an 18‐years retrospective cohort study. Journal of Clinical
Periodontology, 42, 943–950. https​://doi.org/10.1111/jcpe.12460​
Graetz, C., Schwendicke, F., Kahl, M., Dörfer, C. E., Sälzer, S., Springer,
C., … Rühling, A. (2013). Prosthetic rehabilitation of patients with
history of moderate to severe periodontitis: A long‐term evalua‐
tion. Journal of Clinical Periodontology, 40, 799–806. https​ ://doi.
org/10.1111/jcpe.12124​
Haffar, S., Bazerbachi, F., Prokop, L., Watt, K. D., Murad, M. H., & Chari,
S. T. (2017). Frequency and prognosis of acute pancreatitis associ‐
ated with fulminant or non‐fulminant acute hepatitis A: A system‐
atic review. Pancreatology, 17, 166–175. https​ ://doi.org/10.1016/j.
pan.2017.02.008
Higashi, M. K., Veenstra, D. L., del Aguila, M., & Hujoel, P. (2002). The
cost‐effectiveness of interleukin‐1 genetic testing for periodontal
 |
712       HELAL et al.
disease. Journal of Periodontology, 73, 1474–1484. https​ ://doi.
org/10.1902/jop.2002.73.12.1474
Higgins, J. P. T., & Green, S. (2011). Cochrane handbook for systematic re-
views of interventions. Version 5.10 (updated March 2011). London,
UK: The Cochrane Collaboration.
Higgins, J. P. T., & Thompson, S. G. (2002). Quantifying heterogeneity
in a meta‐analysis. Statistics in Medicine, 21, 1539–1558. https​://doi.
org/10.1002/sim.1186
Karimbux, N. Y., Saraiya, V. M., Elangovan, S., Allareddy, V., Kinnunen,
T., Kornman, K. S., & Duff, G. W. (2012). Interleukin‐1 gene polymor‐
phisms and chronic periodontitis in adult whites: A systematic review
and meta‐analysis. Journal of Periodontology, 83, 1407–1419. https​://
doi.org/10.1902/jop.2012.110655
Kassebaum, N. J., Bernabe, E., Dahiya, M., Bhandari, B., Murray, C. J.,
& Marcenes, W. (2014). Global burden of severe periodontitis in
1990‐2010: A systematic review and meta‐regression. Journal of Dental
Research, 93, 1045–1053. https​://doi.org/10.1177/00220​34514​552491
Laster, L., Laudenbach, K. W., & Stoller, N. H. (1975). An evaluation of
clinical tooth mobility measurements. Journal of Periodontology, 46,
603–607. https​://doi.org/10.1902/jop.1975.46.10.603
Lee, C. T., Huang, H. Y., Sun, T. C., & Karimbux, N. (2015). Impact of patient
compliance on tooth loss during supportive periodontal therapy: A
systematic review and meta‐analysis. Journal of Dental Research, 94,
777–786. https​://doi.org/10.1177/00220​3 4515​578910
Leung, W. K., Ng, D. K., Jin, L., & Corbet, E. F. (2006). Tooth loss in treated
periodontitis patients responsible for their supportive care arrange‐
ments. Journal of Clinical Periodontology, 33, 265–275. https​://doi.
org/10.1111/j.1600-051X.2006.00903.x
Listl, S., Galloway, J., Mossey, P. A., & Marcenes, W. (2015). Global eco‐
nomic impact of dental diseases. Journal of Dental Research, 94,
1355–1361. https​://doi.org/10.1177/00220​3 4515​602879
Martinez‐Canut, P. (2015). Predictors of tooth loss due to periodon‐
tal disease in patients following long‐term periodontal mainte‐
nance. Journal of Clinical Periodontology, 42, 1115–1125. https​://doi.
org/10.1111/jcpe.12475​
Moher, D., Liberati, A., & Tetzlaff, J. (2009). Preferred reporting items
for systematic reviews and meta‐analyses: The PRISMA statement.
Annals of Internal Medicine, 151, https​://doi.org/10.7326/0003-
4819-151-4-20090​8180-00135​
Muller, S., Eickholz, P., Reitmeir, P., & Eger, T. (2013). Long‐term tooth loss
in periodontally compromised but treated patients according to the
type of prosthodontic treatment. A retrospective study. Journal of
Oral Rehabilitation, 40, 358–367. https​://doi.org/10.1111/joor.12035​
Muzzi, L., Nieri, M., Cattabriga, M., Rotundo, R., Cairo, F., & Pini Prato, G.
P. (2006). The potential prognostic value of some periodontal factors
for tooth loss: A retrospective multilevel analysis on periodontal pa‐
tients treated and maintained over 10 years. Journal of Periodontology,
77, 2084–2089. https​://doi.org/10.1902/jop.2006.050227
Ng, Y. L., Mann, V., & Gulabivala, K. (2010). Tooth survival following
non‐surgical root canal treatment: A systematic review of the lit‐
erature. International Endodontic Journal, 43, 171–189. https​://doi.
org/10.1111/j.1365-2591.2009.01671.x
Ng, M. C., Ong, M. M., Lim, L. P., Koh, C. G., & Chan, Y. H. (2011). Tooth loss in
compliant and non‐compliant periodontally treated patients: 7 years
after active periodontal therapy. Journal of Clinical Periodontology, 38,
499–508. https​://doi.org/10.1111/j.1600-051X.2011.01708.x
Nibali, L., Farias, B. C., Vajgel, A., Tu, Y. K., & Donos, N. (2013). Tooth
loss in aggressive periodontitis: A systematic review. Journal of
Dental Research, 92, 868–875. https​://doi.org/10.1177/00220​3 4513​
501878
Nibali, L., Zavattini, A., Nagata, K., Di Iorio, A., Lin, G. H., Needleman, I., &
Donos, N. (2016). Tooth loss in molars with and without furcation in‐
volvement ‐ a systematic review and meta‐analysis. Journal of Clinical
Periodontology, 43, 156–166. https​://doi.org/10.1111/jcpe.12497​
Papapanou, P. N., Sanz, M., Buduneli, N., Dietrich, T., Feres, M.,
Fine Daniel, H., … Tonetti Maurizio, S. (2018). Periodontitis:
Consensus report of workgroup 2 of the 2017 World Workshop
on the Classification of Periodontal and Peri‐Implant Diseases and
Conditions. Journal of Clinical Periodontology, 45, S162–S170. https​://
doi.org/10.1111/jcpe.12946​
Ravald, N., & Johansson, C. S. (2012). Tooth loss in periodontally
treated patients: A long‐term study of periodontal disease and
root caries. Journal of Clinical Periodontology, 39, 73–79. https​://doi.
org/10.1111/j.1600-051X.2011.01811.x
Salvi, G. E., Mischler, D. C., Schmidlin, K., Matuliene, G., Pjetursson, B. E.,
Bragger, U., & Lang, N. P. (2014). Risk factors associated with the lon‐
gevity of multi‐rooted teeth. Long‐term outcomes after active and
supportive periodontal therapy. Journal of Clinical Periodontology, 41,
701–707. https​://doi.org/10.1111/jcpe.12266​
Saminsky, M., Halperin‐Sternfeld, M., Machtei, E. E., & Horwitz, J. (2015).
Variables affecting tooth survival and changes in probing depth:
A long‐term follow‐up of periodontitis patients. Journal of Clinical
Periodontology, 42, 513–519. https​://doi.org/10.1111/jcpe.12419​
Schwendicke, F., Biffar, A. S., & Graetz, C. (2017). Long‐term treatment
costs for aggressive periodontitis in a German population. Journal of
Clinical Periodontology, https​://doi.org/10.1111/jcpe.12814​
Schwendicke, F., Graetz, C., Stolpe, M., & Dorfer, C. E. (2014). Retaining
or replacing molars with furcation involvement: A cost‐effectiveness
comparison of different strategies. Journal of Clinical Periodontology,
41, 1090–1097. https​://doi.org/10.1111/jcpe.12315​
Schwendicke, F., Plaumann, A., Stolpe, M., Dorfer, C. E., & Graetz, C.
(2016). Retention costs of periodontally compromised molars in a
German population. Journal of Clinical Periodontology, 43, 261–270.
https​://doi.org/10.1111/jcpe.12509​
Schwendicke, F., Schmietendorf, E., Plaumann, A., Sälzer, S., Dörfer, C.,
& Graetz, C. (2018). Validation of multivariable models for predicting
tooth loss in periodontitis patients. Journal of Clinical Periodontology,
45, 701–710.
Schwendicke, F., Stolpe, M., Plaumann, A., & Graetz, C. (2016). Cost‐ef‐
fectiveness of regular versus irregular supportive periodontal ther‐
apy or tooth removal. Journal of Clinical Periodontology, 43, 940–947.
https​://doi.org/10.1111/jcpe.12595​
Thompson, S. G., & Sharp, S. J. (1999). Explaining heterogeneity in
meta‐analysis: A comparison of methods. Statistics in Medicine, 18,
2693–2708.
Viechtbauer, W. (2010). Conducting meta‐analyses in R with the meta‐
for Package. Journal of Statistical Software, 1(3), 2010. https​://doi.
org/10.18637/​jss.v036.i03
Wells, G. A., Shea, B., O'Connell, D., Peterson, J., Welch, V., & Losos, M.
(2009). The Newcastle‐Ottawa Scale (NOS) for assessing the quality
of nonrandomized studies in meta‐analysis. Retrieved from http://
www.ohri.ca/progr​ams/clini​c al_epide​miolo​g y/nosgen.pdf
S U P P O R T I N G I N FO R M AT I O N
Additional supporting information may be found online in the
Supporting Information section at the end of the article. 
How to cite this article: Helal O, Göstemeyer G, Krois J, Fawzy
El Sayed K, Graetz C, Schwendicke F. Predictors for tooth loss
in periodontitis patients: Systematic review and meta‐analysis.
J Clin Periodontol. 2019;46:699–712. https​://doi.org/10.1111/
jcpe.13118​